Galaasen, Sigu d Møls e ; Raja, Akash
Wo king Pape
The dynamics o s ock ma ke pa icipa ion
Wo king Pape , No. 5/2024
P o ided in Coope a ion wi h:
No ges Bank, Oslo
Sugges ed Ci a ion: Galaasen, Sigu d Møls e ; Raja, Akash (2024) : The dynamics o s ock ma ke
pa icipa ion, Wo king Pape , No. 5/2024, ISBN 978-82-8379-313-0, No ges Bank, Oslo,
h ps://hdl.handle.ne /11250/3172114
This Ve sion is a ailable a :
h ps://hdl.handle.ne /10419/310432
S anda d-Nu zungsbedingungen:
Die Dokumen e au EconS o dü en zu eigenen wissenscha lichen
Zwecken und zum P i a geb auch gespeiche und kopie we den.
Sie dü en die Dokumen e nich ü ö en liche ode komme zielle
Zwecke e iel äl igen, ö en lich auss ellen, ö en lich zugänglich
machen, e eiben ode ande wei ig nu zen.
So e n die Ve asse die Dokumen e un e Open-Con en -Lizenzen
(insbesonde e CC-Lizenzen) zu Ve ügung ges ell haben soll en,
gel en abweichend on diesen Nu zungsbedingungen die in de do
genann en Lizenz gewäh en Nu zungs ech e.
Te ms o use:
Documen s in EconS o may be sa ed and copied o you pe sonal
and schola ly pu poses.
You a e no o copy documen s o public o comme cial pu poses, o
exhibi he documen s publicly, o make hem publicly a ailable on he
in e ne , o o dis ibu e o o he wise use he documen s in public.
I he documen s ha e been made a ailable unde an Open Con en
Licence (especially C ea i e Commons Licences), you may exe cise
u he usage igh s as speci ied in he indica ed licence.
h ps://c ea i ecommons.o g/licenses/by-nc-nd/4.0/
Wo king Pape
The Dynamics o S ock Ma ke Pa icipa ion
No ges Bank Resea ch
Au ho s:
Sigu d Møls e Galaasen
Akash Raja
Keywo ds
Household inance, s ock ma ke
pa icipa ion, dynamics,
expe iences
5 | 2024
No ges Bank Wo king Pape 1
Wo king pape s a No ges Bank, a 1992/1 il 2009/2 kan bes illes på e-
pos : se icesen e @no ges-bank.no
F a 1999 og sene e e publikasjonene ilgjengelige på www.no ges-bank.no
Wo king pape s inneholde o skningsa beide og u edninge som anlig is ikke
ha å sin endelige o m. Hensik en e blan anne a o a e en kan mo a
kommen a e a kollege og and e in e esse e. Synspunk e og konklusjone i
a beidene s å o o a e nes egning.
Wo king pape s om No ges Bank, om 1992/1 o 2009/2 can be o de ed
by e-mail: se icesen e @no ges-bank.no
Wo king pape s om 1999 onwa ds a e a ailable on www.no ges-bank.no
No ges Bank’s wo king pape s p esen esea ch p ojec s and epo s (no
usually in hei inal o m) and a e in ended in e alia o enable he au ho o
bene i om he commen s o colleagues and o he in e es ed pa ies. Views
and conclusions exp essed in wo king pape s a e he esponsibili y o he
au ho s alone.
ISSN 1502-8143 (online)
ISBN 978-82-8379-313-0 (online)
THE DYNAMICS OF STOCK MARKET PARTICIPATION
Sigu d Møls e Galaasen†
No ges Bank
Akash Raja∗
Copenhagen Business School
Danish Finance Ins i u e
No ges Bank
Feb ua y 20, 2024
Abs ac
We documen no el ac s on he exi and een y ma gins o s ock ma ke pa icipa ion by
e ail in es o s using de ailed adminis a i e da a on e e y No wegian esiden om 1993 o
2016. Con a y o he con en ional iew ha indi iduals ei he ne e o always pa icipa e in he
s ock ma ke , we ind ha many households lea e he s ock ma ke wi hin jus 2 yea s o en y.
Such beha io is mo e p ominen o people o low income, weal h, and educa ional a ainmen ,
and hose o younge age. Es ima ion o a haza d unc ion shows ha he e is nega i e du a ion
dependence in exi p obabili ies: he longe households pa icipa e o , he less likely hey a e
o exi . Wi h espec o he een y ma gin, o e 30% o exi e s subsequen ly e u n o he s ock
ma ke , o en jus a yea la e . A s uc u ally-es ima ed li e-cycle model wi h pa icipa ion cos s
ails o gene a e su icien exi s. Ex ending he model o allow o expe ience-based lea ning,
whe eby agen s o m belie s o e he equi y p emium based on hei pe sonal ealized e u ns,
imp o es he model i o pa icipa ion a es, condi ional isky sha es, and inancial weal h- o-
income a ios by o e hal , whils also gene a ing quick exi s and a downwa d-sloping haza d
unc ion o exi . Howe e , he model s ill s uggles o gene a e enough een y. Using g anula
po olio holdings da a, we show ha poo ini ial e u ns a e associa ed wi h quick exi s om he
s ock ma ke , while posi i e e u ns inc ease he likelihood o een y in line wi h an expe ience
e ec s channel.
Keywo ds: household inance, s ock ma ke pa icipa ion, dynamics, expe iences
JEL Classi ica ion: D14, D83, D84, E21, G11, G40, G50
† Bankplassen 2, 0151 Oslo, No way. E-mail: Sigu d-Mols e .Galaasen@No ges-Bank.no
* Co esponding au ho . Solbje g Plads 3, 2000 F ede iksbe g, Denma k. E-mail: [email p o ec ed]
We a e g a e ul o Gual ie o Azzalini (discussan ), F ancesca Bas ianello, Syl ain Ca é (discussan ), Taha
Choukhmane, Gaia Dossi, F ancisco Gomes, Wou e den Haan, Jona hon Hazell, E han Ilze zki, Lu Liu
(discussan ), Jens K ae ne (discussan ), Will Ma cham, Benjamin Moll, Came on Peng, Maa en de Ridde ,
and Rica do Reis, as well as pa icipan s a a ious con e ences and semina s o hei aluable commen s
and sugges ions. Raja acknowledges suppo om he Danish Finance Ins i u e, he Economic and Social
Resea ch Council, and he Paul Woolley Cen e. This pape should no be epo ed as ep esen ing he iews
o No ges Bank. The iews exp essed a e hose o he au ho s and do no necessa ily e lec hose o No ges
Bank.
1
1 In oduc ion
Despi e he la ge a e age e u n on equi ies ela i e o bonds, many households choose
no o in es in he s ock ma ke (Mankiw and Zeldes (1991); Haliassos and Be au (1995);
Campbell (2006)). While he li e a u e has de o ed signi ican a en ion o explaining why
he agg ega e pa icipa ion a e lies below 100%, much less is known abou he mo emen s
in and ou o he s ock ma ke by indi idual in es o s. The con en ional iew is ha indi id-
uals ei he ne e o always in es in he s ock ma ke ; howe e , he da a indica e ha annual
s ock ma ke exi a es a e high, anging om 5-10% in No way. Explo ing he decision o en-
e in o o exi om he s ock ma ke is o i s -o de impo ance because po olio choices
ma e o weal h inequali y (Benhabib e al. (2011); Gabaix e al. (2016); Xa ie (2021)), as
well as he ansmission o mone a y policy (Melcangi and S e k (2020)). Fu he mo e, ana-
lyzing hese ansi ions can help o dis inguish be ween he wide ange o exis ing heo ies o
pa icipa ion ha gene a e agg ega e unde pa icipa ion, gi en ha di e en models ha e
opposing p edic ions o such indi idual-le el mo emen s. In his pape , we shed ligh on
he dynamics o s ock ma ke pa icipa ion by unco e ing no el ac s pe aining o exi and
een y a he indi idual le el using de ailed No wegian adminis a i e da a, and assess he
implica ions o ou indings o heo ies o pa icipa ion.
S udying indi idual-le el changes in pa icipa ion s a us is challenging because panel
da a on weal h holdings o e a long ime dimension a e essen ial. We exploi No wegian
adminis a i e ax eco ds o o e come his challenge. As No way le ies a weal h ax, hese
eco ds con ain de ailed weal h in o ma ion o e e y membe o he popula ion. Ou da a
span 24 yea s, which is signi ican ly longe han simila adminis a i e da ase s om o he
coun ies and o highe equency han mos ele an longi udinal su eys. Indi iduals mus
ile a ax e u n e en i hey hold no inancial asse s, which allows us o con iden ly iden i y
pe iods o nonpa icipa ion. This is a signi ican ad an age ela i e o b oke age accoun s
da a, whe e exi om such samples may simply e lec a ans e o ano he p o ide a he
han a comple e wi hd awal om he s ock ma ke . Financial holdings a e di ec ly epo ed
o he ax au ho i y by he inancial in e media ies hemsel es. Such hi d-pa y epo ing
alle ia es conce ns abou measu emen e o ha can a ise when using sel - epo ed mea-
su es o weal h. In addi ion, we can link he ax eco ds o o he adminis a i e da ase s,
2
he eby gi ing us addi ional in o ma ion abou each ci izen ha is ypically no a ailable in
su ey o b oke age accoun s da ase s.
Using he de ailed No wegian da a, we documen no el ac s on wo ma gins o s ock
ma ke pa icipa ion ha ha e ecei ed less a en ion in he exis ing li e a u e, namely he
exi and een y ma gins. Fi s , we ind ha many indi iduals ha e e y sho spells in he
s ock ma ke ; ha is, hey s ay in he ma ke o only 1–2 yea s and hen comple ely liqui-
da e hei s ock holdings. 20% o all spells end wi hin jus 2 yea s o en y, and his beha io
is no d i en by in olun a y pa icipa ion coming om inhe i ances o employee s ock op-
ions. This nega es he con en ional iew ha once people en e in o he s ock ma ke , hey
should a ely exi . Ou inding he e o e builds on he high exi a es documen ed in Hu s
e al. (1998), Vissing-Jø gensen (2002), and Bonapa e e al. (2023) by showing ha hese
exi s a e pa icula ly d i en by ecen en an s who in es o only a sho pe iod.
We hen in es iga e whe he he likelihood o a sho spell, which is de ined as a spell
ha esul s in comple e exi wi hin 2 yea s, is co ela ed wi h ce ain cha ac e is ics. Cha -
ac e is ics o en linked o lowe inancial li e acy, namely low income, weal h, and educa-
ional a ainmen (Lusa di and Mi chell (2011); Beh man e al. (2012)), a e associa ed wi h
an inc eased likelihood o ha ing a sho spell. We also ind a s ong, posi i e ela ionship
be ween he sha e o inancial weal h in es ed a en y and he likelihood o a sho spell,
meaning his beha io is no d i en by people who ini ially in es small amoun s in o he
s ock ma ke . Wi h ega d o age, sho spells a e mo e common amongs he younges
indi iduals, in line wi h he high exi and en y a es in his age ca ego y documen ed in
Fage eng e al. (2017) and Bonapa e e al. (2023). Men a e 20% mo e likely o exhibi such
beha io compa ed o women, suppo ing exis ing e idence ha men end o ade exces-
si ely and display o e con idence (Ba be and Odean (2001)). Fu he mo e, quick exi s a e
signi ican ly mo e likely among in es o s who en e in o di ec ly held s ocks a he han mu-
ual unds. To es ablish he deg ee o which liquidi y shocks could o ce some in es o s o
lea e he s ock ma ke p ema u ely, we iden i y liquidi y- ela ed e en s in he da a, such as
house pu chases and unemploymen . The e is a posi i e ela ionship be ween expe iencing
a liquidi y need and sho spells, sugges ing ha liquidi y needs can play a ole; howe e ,
app oxima ely 85% o exi e s do no expe ience such e en s a exi , which limi s he deg ee
o which his beha io can be explained pu ely by liquidi y shocks.
3
To unde s and how he p obabili y o exi e ol es wi h ime spen in he ma ke , we
es ima e a haza d unc ion o exi om pa icipa ion using he me hodology o Al a ez
e al. (2021). Thei app oach exploi s he p esence o indi iduals wi h mul iple pa icipa-
ion spells o deal wi h unobse ed he e ogenei y ha , i no p ope ly accoun ed o , can
bias he slope o he haza d unc ion downwa ds (Lancas e (1979)). The haza d unc ion
is ound o be downwa d sloping and con ex, which means ha he longe one s ays in he
s ock ma ke , he less likely hey a e o wi hd aw comple ely om he ma ke . Toge he wi h
he sho spells esul , his inding indica es ha pa icipa ion s a us is pa icula ly agile in
he ini ial yea s ollowing en y.
Mo ing on o he een y ma gin, while mos exi e s do no een e he s ock ma ke , we
ind ha a nonnegligble ac ion do wi h o e 30% o exi e s een e ing wi hin he ollowing
4 yea s. They ypically e u n o he same asse class (mu ual unds o di ec s ockholding)
ha hey p e iously in es ed in. Mos een y occu s soon a e exi , o en jus a yea la e ,
and is mo e likely o high income and weal h indi iduals. We also es ima e a downwa d-
sloping and highly con ex haza d unc ion o een y, implying nega i e du a ion depen-
dence in een y p obabili ies: he longe an indi idual has been away om he s ock ma -
ke , he less likely hey a e o e u n. A e abou a decade o nonpa icipa ion, he likelihood
o een y is e ec i ely ze o.
We hen conside he implica ions o ou empi ical indings o heo ies o s ock ma ke
pa icipa ion. In pa icula , we examine he ex en o which a wo kho se li e-cycle po o-
lio choice model à la Cocco e al. (2005) can p oduce sho - e m dynamics. In his model,
agen s can in es in wo inancial asse s, one isky (s ocks) and he o he sa e (bonds), and
hey ecei e an exogenous labo income in e e y pe iod ha is s ochas ic du ing wo king
li e bu cons an in e i emen . To gene a e a mo i e o nonpa icipa ion, we augmen he
baseline model wi h pa icipa ion cos s, which a e a popula explana ion o limi ed pa ic-
ipa ion in he s ock ma ke .1We conside wo ypes o pa icipa ion cos s: he i s ollows
Gomes and Michaelides (2005) and is an en y cos paid a he s a o a new spell. This cap-
u es he ime and e o spen sea ching o an accoun p o ide o lea ning undamen al
1Unde he co e Cocco e al. (2005) model, he e is ull pa icipa ion a all ages and hus no en y o exi
dynamics. Full pa icipa ion ollows om s anda d po olio heo y, which s a es ha as long as he expec ed
equi y p emium is posi i e, e e yone should in es a leas a small amoun in s ocks (Samuelson (1969); Me -
on (1969,1971)).
4
in es men p inciples. The second is a pe -pe iod pa icipa ion cos paid in e e y pe iod
in which one holds s ocks, and ep esen s he oppo uni y cos o ime needed o moni o
and ebalance one’s in es men po olio e e y pe iod o , al e na i ely, b oke managemen
ees (Vissing-Jø gensen (2002,2003); Fage eng e al. (2017)). We also include a beques mo-
i e o be e ma ch he weal h p o ile du ing e i emen (Gomes and Michaelides (2005);
Fage eng e al. (2017)), as well as a small p obabili y o a la ge s ock ma ke c ash (Fage eng
e al. (2017)) o mo i a e exi , pa icula ly in la e li e.
We s uc u ally es ima e pa ame e s o he model o ma ch li e-cycle p o iles o he s ock
ma ke pa icipa ion a e, condi ional isky sha es, and inancial weal h- o-income a ios
using simula ed me hod o momen s. Simula ed li e-cycle p o iles show ha his model
unde p edic s pa icipa ion in ea ly li e, bu o e p edic s i in la e li e. Agen s in he model
need o build up weal h du ing hei wo king li e o jus i y paying he en y cos , which is
es ima ed o be 1.22% o income (≈$500 in 2011 p ices). Due o he low discoun ac o
o 0.827, which is needed o be e ma ch he weal h p o ile, his akes ime and means he
pa icipa ion a e emains close o ze o un il abou age 40. Following en y, agen s ace a
smalle pe -pe iod cos o 0.29% o income (≈$120) and gene ally ha e an inc easing income
p o ile, leading hem o con inue pa icipa ing un il e i emen , a e which poin hey s a
o exi . I he e o e ollows ha he e a e minimal indi idual-le el dynamics in he model
wi h i ually all agen s con o ming o he “con en ional” iew o pa icipa ion - you ei he
ne e pa icipa e o you always pa icipa e upon en y.
To a ionalize ou empi ical indings, we ex end he model o allow o expe ience-based
lea ning (EBL) à la Malmendie and Nagel (2011), whe eby indi iduals do no know he ue
equi y p emium bu o m belie s based on hei own ealized e u ns. This ing edien is
mo i a ed by he li e a u e on memo y and expe ience e ec s documen ing how pas expe-
iences can ha e long-las ing e ec s on belie s and ac ions.2Consis en wi h Fol yn (2020),
he model wi h EBL mo e han hal es he gap be ween he empi ical a ge momen s and he
model-simula ed momen s. The pa icipa ion a e p o ile shows he la ges imp o emen in
i wi h he ise in pa icipa ion occu ing much ea lie in li e compa ed o he model wi h-
ou EBL. This occu s because we es ima e much lowe pa icipa ion cos s (an en y cos o
2See, o example, G eenwood and Nagel (2009); Malmendie and Nagel (2011,2015); Ande sen e al. (2019);
Bo dalo e al. (2020); B iggs e al. (2021); A ouzi e al. (2023).
5
≈$290 and a pe -pe iod cos o ≈$30), making i mo e appealing o in es in s ocks a an
ea lie age. We also ob ain declines in pa icipa ion a es close o he da a, which is due
o he p esence o EBL. Ou s uc u al es ima ion shows ha agen s ha e ecency bias as in
Malmendie and Nagel (2011). Agen s he e o e pu mo e weigh on ecen e u n ealiza-
ions, which means ha upon expe iencing a poo e u n, hey become mo e pessimis ic
and exi he ma ke , leading o a g adual decline in pa icipa ion a es. Fo condi ional isky
sha es, we also ob ain a be e i . The model wi hou EBL unde p edic s sha es om age
50 onwa ds, pa ly because he highe pa icipa ion cos s mean pa icipan s a e ypically
weal hie and he e o e choose lowe isky sha es (Jaganna han e al. (1996); Cocco e al.
(2005)). Howe e , wi h belie s, he indi iduals who selec in o pa icipa ion a e on a e age
mo e op imis ic and wan a highe isky sha e.
Despi e no being a ge ed in he s uc u al es ima ion, he model wi h EBL is able o
gene a e a dis ibu ion o spell leng hs ha is e y close o he da a. Some indi iduals will
d aw poo e u ns upon en y, making hem mo e pessimis ic abou s ock e u ns going o -
wa d and esul ing in quick exi . We also ob ain a downwa d-sloping haza d unc ion. The
in ui ion o his ollows om he selec ion o who emains in he s ock ma ke . I a pe -
son has emained in he ma ke o a long ime, hey mus ha e expe ienced s ong e u ns,
o he wise he EBL o ce would ha e pushed hem ou o he ma ke . As such, i will equi e
a e y nega i e e u n o make hem su icien ly pessimis ic such ha hey choose o exi
in spi e o he p io good expe iences. Such an e en is o low p obabili y, and he e o e,
he model wi h EBL can gene a e nega i e du a ion dependence in exi p obabili ies. How-
e e , on he een y side, while he model wi h EBL gene a es mo e een y compa ed o he
model wi hou belie s, i is s ill much less han obse ed in he da a wi h jus o e 1% o in-
di iduals expe iencing mul iple spells in he model simula ions compa ed o 12.5% in he
da a. This is because belie s ake ime o eco e unde he es ima ed pa ame e alues, by
which ime indi iduals a e likely o be close o o al eady in e i emen and ha e less desi e
o pa icipa e. Fo his same eason, amongs he se o agen s ha do een e , een y ends
o occu o e a decade a e exi .
We conclude he pape by es ing he expe ience e ec s channel. A he indi idual le el,
we exploi he de ailed in o ma ion on holdings o No wegian lis ed s ocks epo ed in he
Sha eholde Regis y. As we do no obse e in o ma ion on speci ic mu ual und holdings,
6
Figu e 1: S ock ma ke pa icipa ion a e o e ime
0 .1 .2 .3 .4
O e all pa icipa ion a e
1995 2000 2005 2010 2015
Yea
No es. This igu e plo s he pa icipa ion a e in he s ock ma ke annually om 1993 o 2016.
Figu e 2: En y and exi a es o e ime
0 .05 .1 .15 .2
Ra e
1995 2000 2005 2010 2015
Yea
En y a e Exi a e
Oslo s ock ma ke down u n
No es. This igu e plo s he en y and exi a es o s ock ma ke pa icipa ion. The en y a e in yea is he
p opo ion o nonpa icipan s in yea −1 who in es a leas $150 in he s ock ma ke in yea . The exi a e
in yea is he p opo ion o pa icipan s in yea −1 who sell all o hei holdings in yea . The shaded a eas
a e s ock ma ke down u n yea s in which he Oslo Bø s Benchma k Index ell by a leas 10%.
13
3 Empi ical ac s
In his sec ion, we s udy and documen no el ac s pe aining o wo ma gins o s ock ma ke
pa icipa ion using he No wegian adminis a i e da a. Sec ion 3.1 explo es he exi ma gin,
and shows ha sho spells in he s ock ma ke a e common and linked o cha ac e is ics
associa ed wi h low inancial li e acy. Sec ion 3.2 s udies he een y ma gin, whe e we ind
ha a non i ial p opo ion o exi e s do subsequen ly een e he s ock ma ke .
3.1 Exi ma gin
3.1.1 Sho spells a e common
We begin by examining he dis ibu ion o spell leng hs in he da a. Figu e 3plo s a his-
og am wi h he dis ibu ion o spell leng hs based on spells beginning be ween 1994 and
2013 inclusi e.11 We es ic a en ion o spells s a ing by 2013 o ensu e ha pa icipan s
ha e a leas 3 yea s in which o exi . I , o example, 2015 en an s we e also included, hey
would ei he ha e a 1-yea comple ed spell o be igh censo ed. Including such en an s
would he e o e a i icially in la e he ba s co esponding o a sho spell leng h. The his-
og am shows a declining ela ionship be ween spell leng h and he p opo ion o obse a-
ions. Abou 13% o all spells end in jus 1 yea , and 20% end wi hin 2 yea s. We unde ake a
a ie y o obus ness checks, namely analysis a he household le el (Figu e H2), excluding
en an s who ecei e a gi o inhe i ance in he yea o o be o e en y (Figu e H3), emo ing
indi iduals wi h s ocks in he company hey wo k o (Figu e H4), d opping in es o s who
in es a small sum a he poin o en y (Figu e H5), and only using he i s ( eco ded) spell
o each pa icipan (Figu e H6). In all cases, simila pa e ns eme ge.
The nex s ep is unde s anding whe he sho spells can be linked o obse able cha ac-
e is ics. To do his, we es ima e he ollowing linea p obabili y model:
P (spell ends wi hin 2 yea s) =α+δ +β′Xi +ϵi (1)
whe e δ deno es en y-yea ixed e ec s and Xi is a ec o o obse able cha ac e is ics
11Le -censo ed spells a e excluded om his igu e because a spell leng h canno be compu ed o such
spells. These spells a e ypically hose ha we e al eady ongoing a he s a o ou da a, hough o he easons
o le -censo ing la e in he sample could be immig a ion o an exis ing s ockholde in o No way.
14
Figu e 3: Dis ibu ion o spell leng hs
0 .05 .1 .15
F ac ion o spells (n=1822438)
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22
Leng h o spell (in yea s)
No es. This his og am plo s he p opo ion o spells o di e en leng hs in he No wegian da a. We ake all
spells beginning a any poin om 1994 o 2013. The x-axis gi es he spell leng h (in yea s), and he y-axis
shows he p opo ion o spells belonging o a pa icula spell leng h. Righ -censo ed spells a e excluded om
he plo , bu a e used in calcula ing p opo ions.
measu ed a he poin o en y, such as age and weal h.
Table 2shows he coe icien es ima es om his es ima ion.12 Qui e in e es ingly, sho
spells a e no mo e likely amongs hose who pu small amoun s in he s ock ma ke . In-
s ead, inc easing he sha e o inancial weal h in es ed in he s ock ma ke by 10pps in-
c eases he p obabili y o a sho spell by 1.3pps, which is a 6.4% inc ease in he p obabili y
when compa ed o he sample mean o 20%. This implies ha his beha io is no concen-
a ed amongs people who make small gambles. Men a e 25% (4.9pps) mo e likely han
women o ha e a sho spell. This esul ela es o he exis ing li e a u e on gende di -
e ences in con idence and ading beha io , which has ound ha men end o be mo e
o e con iden and ade excessi ely, o en o he de imen o hei own e u ns (Ba be and
Odean (2001)).13 Rega ding age, we see ha sho spells a e mo e likely o he younges age
g oup (Figu e H8). This inding is in line wi h Fage eng e al. (2017), who show ha younge
households end o en e and exi equen ly.14
12We ind simila esul s i a p obi model is used ins ead.
13G inbla and Keloha ju (2009) s udy o e con idence using Finnish da a and show ha indi iduals wi h a
high deg ee o sel -con idence end o ha e highe ading olumes.
14Figu e H11 shows ha sho spells became much mo e likely du ing he ea ly 2000s, a pe iod ha exhibi ed
15
Table 2: De e minan s o sho spells (≤2 yea s)
(1)
Risky sha e (ou o inancial weal h) 0.128∗∗∗
(0.001)
Homeowne -0.019∗∗∗
(0.001)
Male 0.049∗∗∗
(0.001)
Unemployed 0.019∗∗∗
(0.001)
Deg ee -0.010∗∗∗
(0.001)
Single 0.021∗∗∗
(0.001)
Sample mean 0.20
Yea FE Yes
Weal h and income FE Yes
Age g oup FE Yes
Obse a ions 1841303
R-squa ed 0.05
No e: ∗p<0.05, ∗∗ p<0.01, ∗∗∗ p<0.001. This able shows he es ima ion o Equa ion 1. The dependen a iable is a bina y a i-
able equal o 1 i he spell ends wi hin 2 yea s, and ze o o he wise. Risky sha e is he amoun in es ed in public equi y di ided by o al
inancial weal h. Homeowne is a bina y a iable equal o 1 i he pa icipan owns hei own p ope y (ei he sel -owned o owne -
ship h ough housing coope a i es), and ze o o he wise. Single is a bina y a iable equal o 1 i he pa icipan is nei he ma ied no
cohabi ing, and ze o o he wise. Unemployed is a bina y a iable equal o 1 i he pa icipan ecei es unemploymen bene i s a he
poin o en y, and ze o o he wise. En y yea ixed e ec s a e included. Age ixed e ec s by b oad age g oup (20-29, 30-39, 40-49,
50-59, 60-69 and 70+), as well as income and weal h decile ixed e ec s a e included. Obse ables a e measu ed a he poin o en y.
S anda d e o s a e clus e ed a he indi idual le el. The eg ession uses da a on en an s om 1994-2014.
Cha ac e is ics ypically associa ed wi h lowe inancial li e acy (low income, weal h,
and educa ion le els) a e also linked o a highe p e alence o sho spells.15 Ha ing a col-
lege deg ee lowe s he likelihood o exi ing wi hin 2 yea s by abou 5% (1pp). Figu es 4a and
4b show he impac s o income and weal h, espec i ely. We ind a nega i e ela ionship be-
ween income and he p obabili y o a sho spell, wi h hose in he bo om income decile
ha ing a 19% (3.7pps) highe p obabili y o a sho spell ela i e o he median. Fo weal h,
he impac o low weal h is s onge . En an s belonging o he bo om weal h decile a e 28%
(5.5pps) mo e likely o exi wi hin 2 yea s ela i e o he median.16 The ela ionship be ween
signi ican ading olumes and s ock ma ke in- and ou lows associa ed wi h he build-up and subsequen
bu s ing o he do -com bubble (O ek and Richa dson (2003); Hong and S ein (2007)).
15Lusa di and Mi chell (2011) gi e e idence o a posi i e co ela ion be ween educa ional a ainmen and
inancial li e acy. Beh man e al. (2012) ind his as well and u he show a posi i e co ela ion be ween weal h
and inancial li e acy.
16Figu e H7 plo s he weal h dis ibu ion o sho and longe - e m spelle s sepa a ely, and u he shows
16
weal h and he p obabili y o a sho spell is con ex, implying ha he ma ginal impac o
mo e weal h on his p obabili y is declining and e ec i ely becomes ze o by he 7 h decile.
Cal e e al. (2009a) ind ha indi iduals wi h less income, weal h, and educa ion a e mo e
likely o exi . Ou indings he e o e show ha hey a e no jus mo e likely o exi a any
poin . Ra he , hey a e also mo e likely o expe ience a quick exi .
Figu e 4: Impac o income and weal h on he p obabili y o a sho spell
(a) Income
-.01 0 .01 .02 .03 .04
Coe icien es ima e
1 2 3 4 5 6 7 8 9 10
Income decile
(b) Weal h
-.02 0 .02 .04 .06
Coe icien es ima e
1 2 3 4 5 6 7 8 9 10
Financial weal h decile
No es. This igu e plo s he coe icien es ima es o he ixed e ec s on income (a) and inancial weal h (b)
deciles ollowing he es ima ion o Equa ion 1. Va iables a e measu ed a he poin o en y, and deciles a e
based on he ull No wegian popula ion aged 20 and abo e in ha yea . The e ec s a e es ima ed ela i e o
he median g oup. 95% con idence in e als a e shown. The ed line ep esen s a null ela i e e ec .
Finally, we explo e he deg ee o which liquidi y needs could be d i ing hese sho spells.17
In p inciple, some indi iduals may exi he s ock ma ke due o, o example, job loss o un-
o eseen heal h expenses. Upon he “comple ion” o such liquidi y needs, indi iduals may
subsequen ly een e he ma ke . We he e o e exploi he ichness o he No wegian da a
and iden i y liquidi y- ela ed e en s. In pa icula , we look a house pu chases, di o ce, un-
employmen , and a la ge d op in income o >50% as ou liquidi y shocks.18 Figu e H10 plo s
he p opo ion o exi e s o di e en spell leng hs expe iencing a leas one o hese shocks in
ha sho spelle s a e mo e likely o belong o a lowe weal h decile han longe - e m pa icipan s.
17In Appendix B, we p o ide a discussion o o he po en ial explana ions, namely pensions, ma ke iming,
and ax op imiza ion.
18Two o he liquidi y needs could be heal h shocks and educa ion cos s. Howe e , highe educa ion is ee
in No way. While heal hca e is no ee, he e is an annual deduc ible abo e which heal hca e is ee. This
deduc ible is ai ly small a NOK 2,460 in 2021 ($410 in 2011 USD). Ac oss OECD coun ies, No way has he
highes sha e o heal hca e inanced h ough go e nmen schemes and he la ges pe -capi a spending on
heal hca e ela ing o long- e m ca e (Coope (2019)). As such, No wegians in gene al do no seem o be sus-
cep ible o la ge inancial cos s linked o heal hca e needs.
17
hei exi yea . Fo compa ison, we also show he p opo ion o con inuing pa icipan s ex-
pe iencing a liquidi y shock. The e is some link be ween expe iencing a liquidi y shock and
quick exi s as he p e alence o hese liquidi y- ela ed e en s is dec easing in spell leng h.
This is in line wi h pape s linking exi o house pu chases (B andsaas (2021)), ma i al s a us
(Ch is iansen e al. (2015)), and unemploymen (Bas en e al. (2016)). Howe e , i is wo h
no ing ha liquidi y needs a e unable o explain e e y sho spell because i 15% o exi e s
lea e due o one o hese obse ed shocks, i means ha 85% o exi e s a e lea ing o o he
easons.
I is impo an o emphasize ha sho spells a e no exclusi e o ce ain subg oups. Fig-
u e H9 shows he dis ibu ion o spell leng hs by income, weal h, educa ion, gende , and
asse class. Fo example, while men a e mo e likely o exi quickly (Figu e H9d), abou 17%
o women s ill lea e he s ock ma ke wi hin 2 yea s o en y. I is also no ewo hy om Fig-
u e H9e ha he p e alence o sho spells is pa icula ly high o s ock in es o s wi h jus
o e 30% o hei spells ending wi hin 2 yea s compa ed o abou 20% o und in es o s.
Ou inding ha sho spells in he s ock ma ke a e common can ha e impo an impli-
ca ions o weal h accumula ion. Indeed, much o he policy ocus has been on encou aging
en y in o he s ock ma ke (e.g., ia ax incen i es). Howe e , we see ha empo a y pa ic-
ipa ion is e y common, so om a policy pe spec i e, i is no only abou encou aging en y
in o he s ock ma ke . I is also impo an o encou age pa icipan s no o exi impulsi ely
so ha hey can ea n he high equi y p emium on a e age.
3.1.2 Exi p obabili y alls wi h spell du a ion
A e in es o s mo e likely o exi he s ock ma ke in he ini ial pe iods ollowing en y o a -
e s aying in he ma ke o a p olonged pe iod? To answe his ques ion, we es ima e he
haza d unc ion o exi om pa icipa ion. The haza d unc ion hi(d) deno es he p oba-
bili y ha indi idual iexi s he ma ke dyea s a e en y, condi ional on no exi ing un il
hen. A s anda d challenge wi h haza d unc ion es ima ion is sepa a ing ue du a ion de-
pendence om (unobse ed) he e ogenei y. Es ima ing haza d unc ions based on pooled
samples wi h he e ogeneous indi iduals can lead o a downwa d bias in he slope o he haz-
a d unc ion because indi iduals who a e less likely o “su i e” exi he sample ea lie han
o he s (Lancas e (1979); Kie e (1988)).
18
To add ess his conce n, we apply he linea GMM es ima o o Al a ez e al. (2021) and
es ima e a disc e e- ime p opo ional haza d model o du a ion. The main ad an age o
hei app oach is ha i gi es a consis en es ima o o he slope o he haza d unc ion,
e en in he p esence o ime-in a ian indi idual he e ogenei y. Thei me hodology does
so by exploi ing he p esence o indi iduals wi h mul iple spells in he s ock ma ke . The
esul ing limi a ion is ha he se o people expe iencing mul iple spells used in he es i-
ma ion can be undamen ally di e en om he es o he popula ion. Howe e , simila
pa e ns do eme ge when using he ull se o pa icipan s and ins ead es ima ing a Cox p o-
po ional haza ds model (Figu e H12). Fu he de ails on he Al a ez e al. (2021) app oach
a e p o ided in Appendix D.
Figu e 5: Baseline haza d unc ion o exi om pa icipa ion
0 .2 .4 .6 .8 1
Haza d a e (no malized)
0 5 10 15
Yea s since en y
No es. This igu e plo s he es ima ed baseline haza d o exi om pa icipa ion ollowing he me hodology
o Al a ez e al. (2021) desc ibed in Appendix D. The do ed ed lines deno e 95% con idence in e als. The
haza d a e a du a ion d=1 is no malized o 1.
Figu e 5plo s he es ima ed baseline haza d unc ion. The haza d unc ion is mono oni-
cally declining in du a ion, indica ing nega i e du a ion dependence; ha is, he longe one
has been pa icipa ing in he s ock ma ke , he less likely hey a e o exi comple ely a ha
poin in ime. As desc ibed in Appendix D, we a e able o eco e he slope o he baseline
haza d a he han i s le el using he Al a ez e al. (2021) app oach, so we no malize he haz-
a d a e a d=1 o 1. A s iking ea u e o he haza d unc ion is he s eepness o he slope in
he ini ial yea s ollowing en y. The haza d a es a d=2 and d=3 a e abou 55% and 40%
ha o d=1, espec i ely. By d=12, he haza d a e is close o ze o, sugges ing ha i an
19
indi idual emains in he ma ke o a p olonged pe iod, he likelihood o hem comple ely
wi hd awing om he ma ke is minimal. Combined wi h he ac ha many s ock ma ke
pa icipan s s ay in he s ock ma ke o a sho ime, his inding indica es s ong dynamics
in he ini ial yea s ollowing en y.
3.2 Reen y ma gin
3.2.1 Some exi e s een e he s ock ma ke
We now u n o unde s anding whe he exi e s e e een e ollowing exi . Figu e 6plo s
he dis ibu ion o he numbe o spells an indi idual expe iences.19 40% o he popula ion
ne e pa icipa es in s ocks, while 48% a e obse ed o pa icipa e jus once, meaning ha
a ound 12% o he en i e popula ion has mul iple spells. Hence, een y does occu o a
nonnegligble p opo ion o he popula ion. Indeed, his inding nega es he con en ional
iew in he li e a u e ha upon en e ing he s ock ma ke , indi iduals should a ely lea e.
He e, we see ha some people liquida e hei s ock holdings comple ely bu subsequen ly
een e .20 Figu e H15 plo s a ime se ies o he een y a e, and shows ha since he la e
1990s, he een y a e has ho e ed be ween 20-40%. The e o e, while mos exi e s do no
een e , a nonnegligible p opo ion o hem do.
We also ind ha in es o s end o e u n o he same asse class in which hey p e iously
pa icipa ed. Figu e H14 shows ha o e 80% o een an s who p e iously pa icipa ed in
unds choose o e u n o unds. O hose een an s who p e iously in es ed in indi idual
s ocks, o e 60% go back in o di ec s ockholding. This esul sugges s ha in es o s end o
di ide hemsel es in o ypes, namely und in es o s and di ec s ockholde s, wi h e y ew
pa icipa ing in bo h.
We now examine which cha ac e is ics a e associa ed wi h een y. Fo his pu pose, we
un he ollowing linea p obabili y model:
P ( een e wi hin 4 yea s) =α+δ +β′Xi +ϵi (2)
19We es ic a en ion o indi iduals who appea in he da a o a leas 15 yea s, as hose who appea o
ewe yea s a e likely o ha e ei he ze o o one spell, which would skew he dis ibu ion o he le .
20Figu e H13 decomposes he en y a e in o een an s and new en an s, and shows ha abou one- hi d
o one-hal o en an s in a gi en yea a e een an s.
20
Figu e 6: Numbe o spells
0 .1 .2 .3 .4 .5
P opo ion o indi iduals (n=3988853)
0 1 >1
Numbe o pa icipa ion spells
No es. This igu e plo s he dis ibu ion o he numbe o spells using indi iduals who appea in he da a o a
leas 15 yea s.
whe e δ now deno es exi -yea ixed e ec s. We use a ixed window o 4 yea s o een e
because hose who exi ea ly in he sample ha e mo e yea s emaining in which o een e .
A ixed window means all exi e s ha e he same amoun o ime o een e . Fu he mo e,
o p e iew he indings in Sec ion 3.2.2, mos een y occu s soon a e exi , and so a 4-
yea window should cap u e a la ge p opo ion o een y. To ensu e ha all indi iduals a e
obse ed o a leas 4 yea s ollowing exi , we es ic a en ion o hose who lea e he s ock
ma ke by 2012.
Table 3shows he coe icien es ima es. Males a e mo e likely o een e by abou 11%
(3.8pps) ela i e o emales, which again is in line wi h he o e con idence and excessi e
ading beha io o males documen ed in Ba be and Odean (2001). We ind ha he cha -
ac e is ics ound o be posi i ely linked o en y in Cal e e al. (2009a) a e also associa ed
wi h a g ea e likelihood o een y, namely ha ing a college deg ee and being o high in-
come (Figu e H16a) and weal h (Figu e H16b). Reen y is also mos likely o he younges
age g oups (Figu e H17), in line wi h he inding in Fage eng e al. (2017) ha pe manen
exi ises sha ply a e e i emen .
21
Table 3: De e minan s o een y
(1)
Homeowne 0.009∗∗∗
(0.001)
Male 0.038∗∗∗
(0.001)
Unemployed -0.015∗∗∗
(0.002)
Deg ee 0.037∗∗∗
(0.001)
Single -0.032∗∗∗
(0.001)
Sample mean 0.35
Yea FE Yes
Weal h and income FE Yes
Age g oup FE Yes
Obse a ions 1226078
R-squa ed 0.15
No e: ∗p<0.05, ∗∗ p<0.01, ∗∗∗ p<0.001. This able shows he es ima ion o he linea p obabili y model in Equa ion 2. The de-
penden a iable is a bina y a iable equal o 1 i he exi e e-en e s wi hin 4 yea s ollowing exi , and ze o o he wise. Homeowne is a
bina y a iable equal o 1 i he pa icipan owns hei own p ope y (ei he sel -owned o owne ship h ough housing coope a i es),
and ze o o he wise. Single is a bina y a iable equal o 1 i he pa icipan is nei he ma ied no cohabi ing, and ze o o he wise. Un-
employed is a bina y a iable equal o 1 i he pa icipan ecei es unemploymen bene i s a he poin o exi , and ze o o he wise. Exi
yea ixed e ec s a e included. Age ixed e ec s by b oad age g oup (20-29, 30-39, 40-49, 50-59, 60-69 and 70+), as well as income and
weal h decile ixed e ec s a e included. Obse ables a e measu ed a he poin o exi . S anda d e o s a e clus e ed a he indi idual
le el. The eg ession uses da a on exi e s om 1994-2012.
3.2.2 Reen y o en occu s soon a e exi
Condi ional on occu ing, how soon a e exi do indi iduals een e ? Figu e 7plo s he
dis ibu ion o een y imes obse ed in he da a. Almos hal o all een y occu s jus 1
yea a e exi , indica ing ha een y ends o be quick. Combined wi h he e idence o
sho spells gi en in Sec ion 3.1.1, his implies ha he e is u no e be ween pa icipa ion
and nonpa icipa ion s a es, wi h many indi iduals d opping ou o pa icipa ion spells a e
only a ew yea s and some exi e s een e ing soon a e exi . These indings a e obus o
excluding ecipien s o gi s o inhe i ances (Figu e H18) and indi iduals holding s ocks in
he company hey wo k o when hey een e (Figu e H19).
3.2.3 P obabili y o een y alls wi h ime since exi
Ou inal ac s udies how he likelihood o een y changes wi h he du a ion since exi . The
objec o in e es is a haza d unc ion h(d), which deno es he p obabili y o een e ing d
22
ha he e is some e e sion in belie s ollowing a bad e u n ealiza ion wi h he speed o e-
e sion de e mined by λand age. We also assume ha agen s display naï e y by no hink-
ing ha hey will change hei belie s in he u u e when making decisions oday, hence
ˆ
µi, +1=ˆ
µi om he pe spec i e o he indi idual. This simpli ica ion e lec s cogni i e lim-
i a ions o agen s when making such complica ed in es men decisions (Fiske and Taylo
(2013)). I also simpli ies he solu ion o he model as agen s do no need o in e nalize
he belie upda ing p ocess when sol ing hei decision p oblem. No e ha belie s will s ill
e ol e in acco dance wi h Equa ion 9when simula ing he model.
Taken oge he , he op imiza ion p oblem in he ex ended model wi h EBL is as ollows:
V ¡x,
1
(α−1>0), ˆ
µ¢=max
C,αh(1−β)c1−1
ψ+β¡π Ex′V1−γ
+1¡x′,
1
(α>0), ˆ
µ¢+κ(1−π )Ex′x′1−γ¢
1−1
ψ
1−γi1
1−1
ψ
(11)
wi h he same cons ain s and e olu ion o cash on hand as be o e. De ails on he model
solu ion me hod a e p o ided in Appendix F.1.
4.2 Pa ame e iza ion
We employ a wo-s ep app oach o calib a ing he model. In he i s s ep, we do an ex e -
nal calib a ion and se ce ain pa ame e s o ei he exis ing alues in he li e a u e o alues
compu ed di ec ly om ou da a. The second s ep in ol es es ima ing p e e ence pa am-
e e s, pa icipa ion cos s, and pa ame e s o he belie upda ing p ocess using simula ed
me hod o momen s (SMM).
Table 4summa izes he pa ame e alues om ex e nal calib a ion. Indi iduals a e bo n
a age b=20 and die o ce ain a e age d=100. The pa ame e s ela ing o he s ochas ic
componen o he labo income p ocess a e aken om Fage eng e al. (2017), who es ima e
he p ocess speci ically o No way. We es ima e he de e minis ic componen o he income
p ocess ( ) ou sel es using he app oach desc ibed in Appendix F.2. Financial ma ke a i-
ables a e also speci ic o No way. The mean equi y p emium and s anda d de ia ion o s ock
e u ns a e aken om Fage eng e al. (2017), who adjus he alues in Dimson e al. (2008) o
accoun o holdings o o eign equi ies by No wegian e ail in es o s. We use he es ima ed
alue o p ail om Fage eng e al. (2017).26 We also i a Pa e o dis ibu ion o inancial weal h
26Fage eng e al. (2017) a gue ha he equency o ail e en s implied by hei es ima ed p obabili y p ail =
29
o indi iduals aged 20, he eby ob aining scale and shape pa ame e s. In ou simula ions,
we d aw ini ial weal hs om his dis ibu ion.27
This lea es 8 pa ame e s ha need o be es ima ed wi hin he model: he p e e ence
pa ame e s (β,γ,ψ,ζ), pa icipa ion cos s ( ¯
F0,¯
F1), and he belie upda ing pa ame e s (ι,λ).
Deno ing θas he pa ame e ec o o es ima e, SMM compa es simula ed momen s m(θ)
ob ained in he model wi h equi alen empi ical momen s m, and selec s he pa ame e
combina ion ˆ
θ ha minimizes he pe cen age di e ence be ween hem. Mo e o mally:
ˆ
θ=a gmin
θ¡m(θ)−m¢′W¡m(θ)−m¢(12)
whe e Wis a weigh ma ix. We use h ee ca ego ies o a ge momen s, namely he s ock
ma ke pa icipa ion a e, he condi ional isky asse sha e, and inancial weal h- o-income
a ios. We compu e each o hese momen s o e e y age om 20 o 85, lea ing us wi h 198
a ge momen s. Fu he de ails on he SMM p ocedu e a e p o ided in Appendix F.3.
Table 4: Ex e nally-calib a ed pa ame e s
Pa ame e Desc ip ion Value Sou ce
Ins i u ional
Re i emen age 67 No wegian law
π Cond’l su i al p obabili ies - SSB Li e Tables 2010
Labo ma ke
( ) De e minis ic wage p o ile - Own calcula ions (Appendix F.2)
φ e Replacemen a io 0.842 Fage eng e al. (2017)
σzS d. de o pe manen shock 0.110 Fage eng e al. (2017)
σuS d. de o empo a y shock 0.152 Fage eng e al. (2017)
Financial ma ke
R Risk- ee e u n 0.0143 Klo land (2004)
µA e age equi y p emium 0.0314 Fage eng e al. (2017)
σϵS d. de o s ock e u n 0.238 Fage eng e al. (2017)
R ail Tail e en e u n -0.485 Fage eng e al. (2017)
p ail Tail e en p obabili y 0.011 Fage eng e al. (2017)
Ini ial weal h
αshape
x0Pa e o shape (ini ial weal h) 0.881 Own calcula ions
αscale
x0Pa e o scale (ini ial weal h) 2.894 Own calcula ions
No es. This able shows he ex e nally-calib a ed pa ame e alues used in ou model simula ions.
0.011 (1-2 ail e en s o e one’s li e ime) is in line wi h ha implied by his o ical s ock ma ke c ashes in No -
way.
27We ake eal inancial weal h a age 20 and im a he 99 h pe cen ile p io o i ing he Pa e o dis ibu ion.
30
5 Model esul s
In his sec ion, we discuss he es ima ed pa ame e s and model simula ions. We begin wi h
a discussion o a model wi h only pa icipa ion cos s (no belie s) and i s abili y o ma ch he
a ge momen s, as well as ou empi ical indings on indi idual-le el dynamics. We hen
analyze he impac o inco po a ing EBL in Sec ion 5.2, be o e doing es s o he model p e-
dic ions in Sec ion 5.3.
5.1 Model wi hou belie s
Column 3 o Table 5gi es he es ima ed pa ame e alues o a model wi hou EBL, i.e.
whe e pa icipa ion cos s ac as he sole ic ion o pa icipa ing in he s ock ma ke . We
ob ain a discoun ac o o β=0.827. While his alue is low ela i e o hose used in mac o
models o es ima es based on li e cycle models o consump ion-sa ings decisions (e.g., Gou -
inchas and Pa ke (2002)), his alue is in he ange o ypical es ima es om li e cycle mod-
els o po olio choice. Fo example, Fage eng e al. (2017) ob ain es ima es o β anging
om 0.75-0.83, while Coope and Zhu (2016) ge alues be ween 0.76-0.90.28 The low dis-
coun ac o is needed o limi weal h accumula ion such ha he inancial weal h p o ile
o e age is mo e in line wi h he da a. We u he ob ain a mode a e coe icien o ela i e
isk a e sion o 6.581, which is ai ly close o he popula ion a e age alue o 7.57 ound in
Cal e e al. (2021) and also in line wi h exis ing es ima es in he li e cycle po olio choice
li e a u e.29 Ou EIS is ψ=0.443, which, when combined wi h ou es ima e o isk a e sion,
implies ha he powe u ili y es ic ion ha γand ψa e in e sely ela ed does no hold. As
such, ou es ima es lend suppo o he Eps ein-Zin o mula ion. Like in Coope and Zhu
(2016), Fage eng e al. (2017), and B iggs e al. (2021), we ind a s ong beques mo i e ha
will help o slow down asse decumula ion du ing e i emen . Fo pa icipa ion cos s, we
ind en y cos s o 1.22% o pe manen income, which app oxima ely amoun s o ≈$500 in
2011 p ices. This is sligh ly lowe han he 2% ound in Alan (2006). The pe -pe iod pa ici-
28No e ha some pape s ob ain highe βes ima es by adding u he ing edien s o he model. Ca he ine
(2021) conside s cyclical skewness o labo income shocks, and ob ains β alues anging om 0.91-0.96. How-
e e , in a mo e s anda d model wi h pa icipa ion cos s bu no cyclical skewness, hey es ima e lowe discoun
ac o s o 0.67-0.88.
29Choukhmane and de Sil a (2023) no e ha exis ing es ima es in he li e a u e ange om 4-14.4 wi h an
a e age o 7.82.
31
pa ion cos is es ima ed o be 0.29% o pe manen income, which is ≈$120. This is wi hin he
ange o es ima es om Fage eng e al. (2017) o $64-$344, and sligh ly lowe han es ima es
in Vissing-Jø gensen (2002) ($300) and Ca he ine (2021) ($250).
Table 5: In e nally-es ima ed coe icien s
Pa ame e No EBL Wi h EBL
P e e ence pa ame e s
βDiscoun ac o 0.827 0.813
γRisk a e sion 6.581 9.573
ψEIS 0.443 0.599
ζBeques mo i e 4.600 4.330
Pa icipa ion cos s
¯
F0En y cos 1.22% 0.70%
¯
F1Pe -pe iod cos 0.29% 0.07%
Belie upda ing
ιP io weigh - 0.029
λExpe ience weigh - 1.644
# o Momen s 198 198
Objec i e unc ion 29.67 13.83
No es. This able gi es he es ima ed pa ame e alues o he models wi hou (Column 3) and
wi h expe ience-based lea ning (Column 4). Pa ame e alues a e es ima ed using simula ed
me hod o momen s, and he objec i e unc ion is calcula ed as in Equa ion 12. Pa icipa ion
cos s a e o mula ed as a pe cen age o pe manen income (see Sec ion 4.1.2). Fo mo e de ails
on he es ima ion p ocedu e, see Sec ion 4.2 and Appendix F.3.
Figu e 9shows he simula ed li e cycle p o iles o ou a ge ed momen s, namely he
s ock ma ke pa icipa ion a e, condi ional isky asse sha es, and he inancial weal h- o-
income a io. Wi hou belie s, he model unde p edic s pa icipa ion a a young age, bu
o e p edic s pa icipa ion in la e li e (Figu e 9a). This occu s because he cons ain o
en e ing he s ock ma ke is paymen o he en y cos . As indi iduals build up weal h du ing
hei wo king li e, hey s a o c oss he cash on hand h eshold equi ed o en e he s ock
ma ke (Figu e H21). Howe e , his akes ime and he pa icipa ion a e is close o ze o un il
abou age 40. Once indi iduals e en ually s a o en e , hey ace he smalle pe -pe iod
cos o pa icipa ing. As hey a e con inuing o build weal h o e his pe iod (as e lec ed by
he inc easing sa ings a e in Figu e H22) and ace an inc easing income p o ile (Figu e F1),
hey will ypically emain in he s ock ma ke . Pa icip ion hus con inues o inc ease un il
e i emen age, a e which indi iduals will s a decumula ing hei asse holdings (bu no
comple ely due o he beques mo i e). Hence, s ock ma ke pa icipa ion begins o all, bu
32
s ill emains abo e le els in he da a.
Figu e 9: Li e cycle p o ile o pa icipa ion, isky sha es, and weal h ( a ge ed momen s)
(a) Pa icipa ion a e
20 30 40 50 60 70 80 90
Age
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
Pa icipa ion a e
Da a
Model w/o belie s
Model wi h belie s
(b) Condi ional isky sha e
20 30 40 50 60 70 80 90
Age
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
Condi ional isky sha e
Da a
Model w/o belie s
Model wi h belie s
(c) Financial weal h- o-income a io
20 30 40 50 60 70 80 90
Age
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
2.2
2.4
Financial weal h o income a io
Da a
Model w/o belie s
Model wi h belie s
No es. This igu e plo s he a e age li e cycle p o iles o he pa icipa ion a e (panel a), he condi ional isky
asse sha e (panel b), and he inancial weal h- o-income a io (panel c) om he No wegian da a, he model
wi hou belie s, and he model wi h EBL.
Figu e 9b shows he li e cycle p o ile o he condi ional isky sha e. A e age condi ional
sha es ini ially ise in he i s decade o li e, hough pa icipa ion a es a e essen ially ze o
in his ime. F om age 40, a e age isky sha es decline. The in ui ion o his ollows om Ja-
ganna han e al. (1996), Cocco e al. (2005), and Gomes (2020). Labo income can be hough
o as an implici holding o he iskless bond gi en ha labo income is a close subs i u e
o bonds han s ocks (Hea on and Lucas (1997)). O e wo king age, he a io o human cap-
i al (p esen alue o u u e labo income) and inancial weal h declines as indi iduals build
33
mo e weal h and you ge close o e i emen , leading indi iduals o il hei po olios away
om s ocks. Beyond e i emen , whe he isky sha es ise o all wi h age depends on he
a e by which human capi al alls ela i e o inancial weal h. Unde ou calib a ion, hu-
man capi al alls a a as e a e han inancial weal h because indi iduals ha e a beques
mo i e ha encou ages hem o con inue sa ing in e i emen . Hence, a e age isky sha es
con inue o all wi h age.
The li e cycle p o ile o weal h (Figu e 9c) shows e y simila pa e ns o he da a. A
low discoun ac o βis needed, o he wise agen s in he model will accumula e oo much
weal h ela i e o he da a. Indi iduals accumula e weal h du ing wo king age o p ecau-
iona y easons gi en ha labo income isk is uninsu able, as well as o e i emen because
a e age incomes in e i emen a e less han in wo king li e. The e is a sha p spike jus be-
o e e i emen age. This e lec s unce ain y abou he le el o pe manen income ha will
p e ail du ing e i emen . Following e i emen , weal h ini ially declines, pa ly e lec ing
he esolu ion o unce ain y o e income, bu also a educ ion in a e age incomes du ing
e i emen ha means agen s sa e less. Howe e , highe mo ali y and he p esence o a
beques mo i e lead people o sa e and weal h o g adually inc ease once again.
We now s udy whe he he pa icipa ion cos model is able o gene a e he indi idual-
le el dynamics ound in he empi ical analysis.30 Figu e 10 plo s he dis ibu ion o spell
leng hs in he simula ed model compa ed o he da a. Unsu p isingly gi en he discussion
on pa icipa ion a es, he es ima ed model ails o gene a e sho spells. Ins ead, once peo-
ple en e , hey gene ally do no exi un il la e li e. Mos spells hus end up as being igh -
censo ed. This is u he ei e a ed in Figu e 11, which shows ha i ually no-one has mul-
iple spells in he s ock ma ke .
5.2 Model wi h belie s
We now conside he model wi h EBL. Column 4 o Table 5shows he es ima ed pa ame e
alues. While he discoun ac o βand he beques mo i e s eng h ζha dly change, he e
is a sha p inc ease in isk a e sion om γ=6.581 o γ=9.573. This is needed because wi h
30In his analysis, we censo he simula ed agen s in he same way as in he da a. Fo example, someone bo n
in 1990 would only appea in ou da a om 2010-2016 gi en he lowe bound on age o 20 in he empi ical
analysis. As such, only hei beha io om ages 20-26 will be used wi h he emaining ages ea ed as censo ed
obse a ions. The dis ibu ion o coho s in he model is se o ma ch ha o he No wegian da a.
34
Figu e 10: Dis ibu ion o spell leng hs
0 5 10 15 20 25
Spell leng h (yea s)
0
0.05
0.1
0.15
P opo ion o spells
Da a
Model w/o belie s
Model wi h belie s
No es. This igu e plo s he dis ibu ion o spell leng hs in he da a (black line) compa ed o he models wi hou
(blue) and wi h ( ed) belie s. Righ -censo ed spells a e excluded om he plo , bu a e used in calcula ing
p opo ions.
Figu e 11: Numbe o spells dis ibu ion
0 1 >1
Numbe o spells
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
P opo ion o indi iduals
Da a
Model w/o belie s
Model wi h belie s
No es. This igu e plo s he dis ibu ion o he numbe o spells in he da a (black) compa ed o he models
wi hou (blue) and wi h ( ed) belie s. In he model simula ions, we es ic a en ion only o hose agen s who
appea in he censo ed sample o a leas 15 yea s o ma ch he app oach used in he empi ical analysis.
35
belie s, he indi iduals who will choose o pa icipa e in he s ock ma ke will on a e age
be mo e op imis ic. This op imism o e he equi y p emium leads hem o choose a g ea e
isky sha e. The highe deg ee o isk a e sion hus helps o dampen his inc ease in a e age
sha es and keeps hem mo e in line wi h he da a. We see ha pa icipa ion cos s all sha ply.
The en y cos almos hal es om 1.22% o 0.7% o pe manen income, while he pe -pe iod
cos alls by h ee qua e s om 0.29% o 0.07%. This occu s because adding belie s p o ides
ano he ic ion ha can gene a e nonpa icipa ion, namely pessimis ic belie s. As such, he
model does no need o load as hea ily on pa icipa ion cos s o keep people ou o he s ock
ma ke .
The EBL model adds wo new pa ame e s o he model, namely he p io weigh ιand he
expe ience weigh λ. Wi h ι=0.029, he p io belie has a long-las ing e ec on belie s. While
he weigh on he p io declines wi h age, i s ill emains high e en in la e li e (Figu e H23).
We ind an expe ience weigh o λ=1.644, which sugges s ha indi iduals ha e ecency
bias and o e weigh ecen pe sonal expe iences. This alue o λlies wi hin he ange o
es ima es by Malmendie and Nagel (2011) o 1.3-1.9.
In e ms o he a ge ed momen s, he model wi h EBL p o ides an imp o ed i o he
a ge ed momen s wi h he objec i e unc ion mo e han hal ing om 29.67 o 13.83.31 Fo
pa icipa ion (Figu e 9a), we ob ain a quicke ise in pa icipa ion wi h a sha p inc ease in
en y om a ound age 25. This occu s o wo easons: i s , he lowe pa icipa ion cos s
means less weal h is needed o jus i y paying he cos s, making i mo e appealing o in es .
Second, he highe deg ee o isk a e sion encou ages mo e p ecau iona y sa ing as indi-
iduals ha e g ea e p udence, hough mo e isk a e sion can also b ing abou he opposi e
e ec by lowe ing op imal isky sha es. Howe e , he low pa icipa ion cos s can make i
s ill wo hwhile o in es , e en when you selec a low sha e. The pa icipa ion a e does
peak ea lie compa ed o he da a, which e lec s he impac o bad expe iences. An ad e se
e u n ealiza ion makes indi iduals pessimis ic, an e ec ha is ampli ied due o ecency
bias. Consequen ly, some indi iduals will exi as soon as hey expe ience a poo e u n,
leading o a gene al decline in pa icipa ion. The pa icipa ion a e du ing e i emen does,
howe e , i mo e closely now han wi hou belie s.
31Fol yn (2020) inds ha inco po a ing EBL in o a li e cycle po olio choice model can close hal o he gap
be ween model-gene a ed pa icipa ion a es and ue a es in he da a, hough hey ind ha he i o a e age
condi ional isky sha es does no imp o e.
36
Fo condi ional isky sha es (Figu e 9b), he model acks he da a be e han in he p e-
ious model, o he han a he e y ea ly pa o li e. The la ge inc ease in a e age sha es
du ing ea ly li e is simply because hose who choose o con inue pa icipa ing mus ha e
had a good e u n ealiza ion. Gi en he limi ed expe iences ha hey ha e had hus a , his
pa icula ealiza ion is gi en a la ge weigh , hus gene a ing e y high isky sha es. The e-
a e , we see ha a e age isky sha es a e ypically la ge han hose ound in he model
wi hou belie s. Again, his is due o selec ion: hose who choose o pa icipa e a e, on a e -
age, mo e op imis ic and hus choose highe isky sha es.
While he model wi h EBL does ha e he quali a i e pa e n o he empi ical li e cycle
p o ile o inancial weal h- o-income a ios, he quan i a i e i is sligh ly weake compa ed
o a model wi h only pa icipa ion cos s. This di e ence is mos s a k om he end o wo k-
ing li e, and e lec s he impac o bad expe iences gene a ing some pe manen exi and
hese agen s consequen ly ob aining a lowe e u n o hei sa ings.
Tu ning o indi idual-le el dynamics, which a e no a ge ed in ou es ima ion, we ind
ha he model wi h EBL gene a es easonable en y and exi a es o 1.85% and 5.77%, e-
spec i ely. These numbe s a e in line wi h he a es ound o No way in Figu e 2. Figu e 10
shows ha he model wi h EBL is able o p oduce a spell leng h dis ibu ion easonably close
o he da a. Sho spells occu because some indi iduals will d aw poo ini ial e u ns, which
lowe s he expec ed e u n on s ocks. The p esence o pa icipa ion cos s can gene a e an
addi ional ma gin o exi by u he educing he ne gain om s ock ma ke pa icipa ion.
As such, ollowing a poo e u n, he h eshold weal h an indi idual needs o con inue in-
es ing inc eases, which may d i e some indi iduals ou o he ma ke e en i hey belie e
ha s ocks will ou pe o m bonds on a e age. Howe e , he e y small pe -pe iod pa ici-
pa ion cos s mean his ampli ica ion is weak.
We a e also able o gene a e a downwa d-sloping haza d unc ion (Figu e 12). As ime
spen in he s ock ma ke inc eases, he ac ha he agen has no ye le he ma ke mus
mean ha hey pe o med well in hei spell hus a , and he e o e, hey should be op i-
mis ic abou he equi y p emium. Consequen ly, one equi es a e y low e u n o undo his
con idence and be d i en ou o he ma ke .
Figu e 11 shows ha he model wi h EBL gene a es a di e en dis ibu ion o numbe o
spells. The e a e much ewe ne e pa icipan s in his model because pa icipa ion cos s
37
Figu e 12: Haza d unc ion o exi
0 2 4 6 8 10 12
Yea s since en y
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Haza d a e (no malized)
Da a
Model wi h belie s
No es. This igu e plo s he haza d a e o exi unde he model wi h belie s ( ed) and he da a (black). The
haza d a e a 1 yea a e en y is no malized o 1 o acili a e compa ison wi h he empi ical haza d unc ion,
o which only he slope is iden i ied.
a e much lowe . Gi en ha indi iduals s a wi h a p io belie equal o he ue equi y p e-
mium and hey expe ience his alue when no pa icipa ing in he s ock ma ke , agen s
ha e a belie o µ=3.14% un il hei i s pa icipa ion spell. Consequen ly, he only ic ion
o pa icipa ing o a i s ime is he pa icipa ion cos s, and so hei sha p educ ion leads
o much ewe ne e pa icipan s. The coun e o his is ha he e a e many mo e agen s
who expe ience one spell. In e ms o een y, he model wi h EBL does gene a e mo e mul-
iple spelle s ela i e o he model wi h only pa icipa ion cos s, o which e ec i ely no-one
een e s, bu he amoun o een y emains much lowe wi h jus 1.2% o indi iduals expe-
iencing mul iple spells compa ed o 12.5% in he da a. The eason o his is ha while
he e is some e e sion in belie s o e ime because agen s expe ience he a e age ma ke
e u n whene e hey a e no pa icipa ing, unde he es ima ed λpa ame e , he speed o
his e e sion is no as enough o b ing exi e s back in o he s ock ma ke quickly.32
32This is illus a ed in Figu e H24, which plo s he e olu ion o belie s o an indi idual who en e s he s ock
ma ke a age 25, expe iences a -20% e u n, and hen pe manen ly exi s he ma ke . By he ime belie s ha e
“ eco e ed”, agen s a e ypically in he phase o li e whe e hey will no ind i wo hwhile o pay he en y cos s
gi en hei cash on hand a ha ime. Plo ing he dis ibu ion o een y imes in Figu e H25 con i ms ha
mos een y akes place a leas a decade a e exi , so we do no obse e quick een y as in he da a.
38
MERTON, R. C. (1969): “Li e ime Po olio Selec ion unde Unce ain y: The Con inuous-Time Case,”
The Re iew o Economics and S a is ics, 51(3), 247–257.
MERTON, R. C. (1971): “Op imum consump ion and po olio ules in a con inuous- ime model,”
Jou nal o Economic Theo y, 3(4), 373–413.
NAKOV, A., AND G. NUÑO (2015): “Lea ning om expe ience in he s ock ma ke ,” Jou nal o Eco-
nomic Dynamics and Con ol, 52, 224–239.
OFEK, E., AND M. RICHARDSON (2003): “Do Com Mania: The Rise and Fall o In e ne S ock P ices,”
The Jou nal o Finance, 58(3), 1113–1137.
POTERBA, J. M., AND A. A. SAMWICK (1997): “Household Po olio Alloca ion O e he Li e Cycle,”
Wo king Pape 6185, Na ional Bu eau o Economic Resea ch.
SAMUELSON, P. A. (1969): “Li e ime Po olio Selec ion By Dynamic S ochas ic P og amming,” The
Re iew o Economics and S a is ics, 51(3), 239–246.
TVERSKY, A., AND D. KAHNEMAN (1973): “A ailabili y: A heu is ic o judging equency and p oba-
bili y,” Cogni i e Psychology, 5(2), 207–232.
VISSING-JØRGENSEN, A. (2002): “Towa ds an Explana ion o Household Po olio Choice He e ogene-
i y: Non inancial Income and Pa icipa ion Cos S uc u es,” Wo king Pape 8884, Na ional Bu eau
o Economic Resea ch.
(2003): “Pe spec i es on Beha io al Finance: Does "I a ionali y" Disappea wi h Weal h?
E idence om Expec a ions and Ac ions,” NBER Mac oeconomics Annual, 18, 139–194.
XAVIER, I. (2021): “Weal h inequali y in he US: he ole o he e ogeneous e u ns,” A ailable a SSRN
3915439.
45
Online Appendix
Con en s
A Va iable cons uc ion 2
B Po en ial explana ions o sho spells 3
B.1 Sophis ica ed ma ke iming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
B.2 Pensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
B.3 Tax op imiza ion .................................... 4
C The No wegian pension sys em 5
D Fu he de ails on he Al a ez e al. (2021) GMM es ima o 6
E Al e na i e heo ies o pa icipa ion 8
E.1 Nons anda d p e e ences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
E.2 Risks aced by households . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
E.3 Cul u al and social en i onmen . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
F Model solu ion and es ima ion 11
F.1 Solu ion me hod . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
F.2 Es ima ion o li e cycle p o iles in he No wegian da a . . . . . . . . . . . . . . 12
F.3 SMM es ima ion p ocedu e . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
G Compu ing e u ns using he Sha eholde Regis y 14
H Addi ional ables and igu es 15
1
A Va iable cons uc ion
He e, we desc ibe he s eps unde aken o ansla e he ax eco ds in o consis en measu es
o weal h by b oad asse class. TR x.y deno es i em x.y in he ax eco ds based on 2016 i em
codings by he No wegian Tax Adminis a ion (Ska ee a en). No e ha while ax alues a e
epo ed in he aw da a, we ansla e hese alues in o ma ke alues o ou analysis. Fo
inancial weal h, we c ea e he ollowing subclasses:
• Cash and deposi s a e compu ed as he sum o deposi s in No wegian banks (TR 4.1.1),
cash (TR 4.1.3), and deposi s in o eign banks (TR 4.1.9).
• Di ec ly held lis ed s ocks a e gi en by he alue o lis ed No wegian sha es and equi y
ce i ica es, bonds, e c. in he No wegian Cen al Secu i ies Deposi o y (TR 4.1.7).
• Di ec ly held unlis ed s ocks a e gi en by capi al in unlis ed sha es, sha e sa ings ac-
coun s, and secu i ies no lis ed in he No wegian Cen al Secu i ies Deposi o y (TR
4.1.8).
• S ock mu ual und holdings a e gi en by he alue o he sha e componen in holdings
o secu i ies unds (TR 4.1.4).
• Money ma ke /bond unds a e gi en by he alue o he in e es componen in hold-
ings o secu i ies unds (TR 4.1.5).
• Financial weal h held ab oad is gi en by o he axable capi al ab oad such as o eign
sha es, ou s anding claims, bonds, and endowmen insu ance (TR 4.6.2).
• O he inancial asse s a e he sum o ou s anding ecei ables in No way (TR 4.1.6), he
sha e o capi al in housing coope a i es o join ly-owned p ope y (TR 4.5.3), pension
insu ance and li e insu ance (TR 4.5.1 + TR 4.5.2), and o he axable capi al, such as
c yp ocu ency (TR 4.5.4).
Real weal h can be decomposed in o he ollow:
• Housing weal h is he sum o housing owned h ough housing coope a i es (TR 4.3.2.2)
and sel -owned p ope y (TR 4.3.2.1 + TR 4.3.2.3).
2
• O he eal weal h is he sum o boa s (TR 4.2.4), ca s (TR 4.2.5), ca a ans (TR 4.2.6),
holiday homes (TR 4.3.3.1 + TR 4.3.2.3), o he eal es a e (TR 4.3.4 + TR 4.3.5 + TR
4.3.2.3), home con en s and mo able p ope y (TR 4.2.3), ix u es and o he business
asse s (TR 4.4.1 + TR 4.4.2 + TR 4.4.3 + TR 4.4.4), and eal weal h ab oad (TR 4.6.1 + TR
4.3.6.1).
B Po en ial explana ions o sho spells
B.1 Sophis ica ed ma ke iming
Could he sho -li ed en y and exi obse ed in he da a be d i en by sophis ica ed ma -
ke ime s? Pe haps hese indi iduals pu sue sho - e m in es men s a egies and een e
whene e a p omising in es men oppo uni y a ises. I his we e he case, we would expec
sho spelling o be co ela ed wi h p oxies o inancial sophis ica ion. Howe e , as dis-
cussed in Sec ion 3.1.1, sho spelling is nega i ely co ela ed wi h cha ac e is ics ypically
associa ed wi h highe inancial li e acy (college educa ion, income and weal h). Fu he -
mo e, we migh expec highe e u ns o mo e sophis ica ed in es o s. Howe e , in Sec ion
5.3, we show ha indi iduals who ha e a sho spell ha e on a e age poo e ini ial e u ns
compa ed o hose who s ay in he ma ke o longe .
B.2 Pensions
One may wo y ha he exis ence o pension weal h could a ec indi iduals’ desi e o ac-
i ely in es in he s ock ma ke ou o hei nonpension weal h. In p inciple, a a ional agen
should conside hei o e all po olio, comp ising bo h pension and nonpension weal h,
when deciding upon hei op imal po olio alloca ion. I , o example, one’s pension weal h
is al eady in es ed in he s ock ma ke , hey may in es less (o no hing a all) ou o non e-
i emen weal h. The e o e, nonpa icipa ion ou o nonpension weal h could simply be a
a ional choice gi en exis ing exposu e h ough pensions.
I pensions a e o be able o explain he dynamics, he ollowing would need o be he
case: 1) he desi ed isky asse sha e ou o o al weal h changes, and indi iduals adjus
hei nonpension holdings o achie e his new goal, and/o 2) exposu e o he s ock ma -
3
ke coming om pension weal h is changing a a high equency, and indi iduals iden i y
hese changes and adjus hei po olio acco dingly. Explaining equen exi and ( e)en y
h ough his ebalancing channel is a guably di icul , as i equi es indi iduals o egula ly
ollow mo emen s in hei pension holdings and o ac i ely ebalance acco dingly. Howe e ,
a ious pape s ha e shown ha po olio adjus men s a e sluggish in bo h e i emen and
non e i emen accoun s (Agnew e al. (2003); Ame iks and Zeldes (2004); B unne meie and
Nagel (2008); Cal e e al. (2009); Ka lsson e al. (2009)). In Appendix C, we p o ide a discus-
sion o he No wegian pension sys em and a gue ha he na u e o he sys em is such ha
pensions a e unlikely o explain he beha io s we obse e.
B.3 Tax op imiza ion
Could he quick exi and een y om he s ock ma ke be due o ax op imiza ion? Pe haps
indi iduals choose o exi in o de o educe hei ax liabili y in a gi en yea . The e a e wo
ax ma gins ha could be ele an he e. The i s is he weal h ax, whe eby indi iduals a e
axed on ne weal h abo e a gi en h eshold.1Howe e , he majo i y o No wegians do no
each he h eshold. This is pa ly due o a o able ax ea men s on ce ain asse classes.
Fo example, he ax alue on housing is 25% o i s ma ke alue. S ocks and mu ual und
holdings a e gi en a alua ion discoun o 45% (in 2021), whe eas cash and deposi accoun
holdings a e no gi en a discoun . I is he e o e ac ually be e o weal h ax pu poses o
e ain weal h in s ocks and unds a he han liquida ing and holding cash. Consequen ly, i
is e y unlikely ha weal h ax conside a ions can explain en y and exi decisions o mos
No wegian households. The second ele an ax is capi al gains ax. In No way, losses made
om he sale o s ocks and equi y unds a e ax-deduc ible, while gains abo e a isk- ee
e u n a e axed. One migh be wo ied ha he quick exi we obse e is because indi idu-
als a e liquida ing hei loss-making sha es o educe hei ax liabili ies.2Howe e , capi al
gains axa ion in No way is ied o he ealiza ion o each indi idual secu i y, no he pe o -
mance o he o e all po olio. To explain he comple e exi ha we obse e, we would need
o see e e y secu i y in one’s po olio making a loss. In addi ion, i ax incen i es a e d i -
1In 2021, ne weal h abo e 1.5m NOK (≈$250,000 in 2011 USD) was axed a 0.85% (0.7% o he municipali y
and 0.15% o he s a e). The h eshold is doubled o couples.
2Using US da a, Odean (1998) shows ha he p e alence o selling losing s ocks is highes in Decembe ,
which can be linked o he end o he ax yea and a emp s o educe ax liabili y.
4
ing his beha io , we migh expec o see een an s pu chasing he same s ock when hey
e u n. While we do no obse e speci ic mu ual und holdings, he Sha eholde Regis y
p o ides in o ma ion on di ec s ock owne ship om 2004. We ind ha only 28% o di ec ly
held s ocks owned jus be o e exi a e hen epu chased upon een y, meaning mos een-
an s a e pu chasing di e en secu i ies. The e o e, we a gue ha ax-mo i a ed selling is
unlikely o d i e ou esul s.
C The No wegian pension sys em
The e a e h ee main componen s o he No wegian pension sys em: he i s is he Na ional
Insu ance Scheme (“ olke ygden”), which is he basic public pension scheme. I ensu es
ha e e yone ecei es a minimum pension income. Fu he mo e, wo ke s a e gua an eed a
supplemen ha is p opo ional o hei income du ing wo king age.3The sys em is de ined-
bene i in na u e, so ci izens ace no s ock ma ke exposu e h ough i . As such, he decisions
o exi and en e he s ock ma ke canno be a ibu ed o po olio ebalancing be ween
p i a e accoun s and public pension weal h.
Second, he e a e occupa ional pensions. Public occupa ional pensions a e also de ined-
bene i schemes. Hence, he e is no s ock ma ke exposu e h ough hem.4P i a e sec o
occupa ional pensions ope a e di e en ly. Un il 2001, only de ined-bene i pensions ex-
is ed. While de ined-con ibu ion pensions, o which he pension bene i depends on how
well he con ibu ions a e in es ed, ha e been allowed since 2001, hey did no gain momen-
um un il 2006 when occupa ional pensions we e made manda o y by law. Indeed, be o e
2006 occupa ional pensions we e mainly p o ided by la ge employe s (OECD (2009)).5One
3Unde he cu en sys em, 18.1% o wages in each yea o employmen up o a ce ain ceiling is ans e ed
o a pension accoun . This pension income is hen indexed o nominal wage g ow h. Upon e i emen , he
accumula ed amoun is no gi en as a lump sum. Ins ead, an annual sum is gi en based on he expec ed num-
be o yea s o be spen as a pensione , which i sel depends on when he indi idual s a s wi hd awing om
hei pension and li e expec ancy. While he e a e some di e ences based on yea o bi h, he o e all p emise
o pensionable income being linked o employmen ea nings s ill holds. Fo u he de ails, see Fage eng e al.
(2019) and F ed iksen and Hal o sen (2019).
4Un il 2020, he public occupa ional pension scheme was such ha wo ke s we e en i led o he maximum
pension a e 30 yea s o se ice and can ecei e a pension equal o 66% o hei pension base ( inal sala y
con e ed in o a ull- ime equi alen ) be o e adjus men s o li e expec ancy. Howe e , occupa ional pension
ea nings became simila o ha in he Na ional Insu ance Scheme om 2020, in pa icula ha ing a sha e o
ea nings each yea be accumula ed in a pension po . Howe e , his emained a de ined-bene i sys em. Fo
u he de ails on public occupa ional pensions and he e o ms, see F ed iksen and S ølen (2018).
5As o 2018, 90% o p i a e sec o employees a e unde a de ined-con ibu ion pension (F ed iksen and
5
may be conce ned ha because p i a e sec o de ined-con ibu ion occupa ional pensions
ha e some exposu e o he s ock ma ke , his could in luence choices made in non e i e-
men in es men accoun s. Howe e , Figu e H11 shows ha sho spells in he s ock ma ke
a e no exclusi e o he pos -2006 pe iod.
Thi d, indi iduals may ha e pe sonal p i a e pensions ha hey in es in. As paymen s
in o an Indi idual Pension Scheme (IPS) in No way a e ax deduc ible up o a ce ain limi ,
one can in e om he ax eco ds whe he an indi idual holds such pensions.6Figu e C1
p o ides a ime se ies o pa icipa ion in p i a e pension accoun s sepa a ely o he whole
popula ion and he subse o he popula ion aged 60 and unde (who a e unlikely o ha e
d awn om such pensions ye ). In ei he case, he pa icipa ion a es a e in single digi s,
indica ing ha he as majo i y o he popula ion do no hold such accoun s. To u he
ease conce ns, we plo he p opo ion o exi e s o di e en spell leng hs who hold p i a e
pensions as o hei exi yea . I hese schemes we e d i ing ou sho spell esul , we migh
expec o see a g ea e p e alence o p i a e pensions among sho spelle s. Howe e , Figu e
H30 shows he opposi e. We also ep oduce ou spell leng h his og am bu exclude any
indi idual who a any poin in he sample holds a p i a e pension accoun . Figu e H31
shows ha ou esul s a e obus o his. We he e o e belie e ha pension holdings canno
explain he sho - e m dynamics we obse e.
D Fu he de ails on he Al a ez e al. (2021) GMM es ima o
The Al a ez e al. (2021) GMM es ima o is based on he ollowing en i onmen . The e is
a p opo ional haza ds da a gene a ing p ocess o du a ions d∈{¯
D,..., ¯
D}, whe e hi(d)=
θibd.θiis he ime-in a ian ail y pa ame e speci ic o indi idual iand cap u es indi id-
ual he e ogenei y in haza d a es. bdis he baseline haza d a du a ion dand is assumed o
be common ac oss indi iduals. The objec i e is o ob ain an es ima e o bd, as his e lec s
ue du a ion dependence a he han unobse ed he e ogenei y. Indi idual iexpe iences
Hal o sen (2019)).
6The e a e wo ele an a iables in he ax da a. TR 3.3.5 eco ds he deduc ible amoun om paymen s
in o an IPS, while TR 4.5.1 indica es capi al in an Indi idual Pension Accoun (IPA). No e ha IPAs we e e-
placed by he IPS in 2006, om which poin new money could no be placed in o one’s exis ing IPA, and new
IPAs could no be opened. We conside an indi idual o be a p i a e pension con ibu o i hey epo a posi-
i e alue o ei he o hese wo a iables, ei he in he cu en yea o in any pas yea .
6
Figu e C1: Pa icipa ion in p i a e pensions o e ime
0 .05 .1 .15
Pa icipa ion in p i a e pensions
1995 2000 2005 2010 2015
Yea
Whole popula ion Popula ion aged ≤60
No es. This igu e plo s a ime se ies o pa icipa ion in p i a e pensions o e ime. The blue line gi es he
pa icipa ion a e o he whole popula ion, while he ed line es ic s a en ion o hose aged 60 o unde .
An indi idual is said o be pa icipa ing in p i a e pensions in a gi en yea i hey pu money in o a p i a e
pension ei he in he cu en yea o in a pas yea . Pa icipa ion has occu ed i ei he o he ollowing wo
i ems in he ax eco ds is nonze o: TR 3.3.5, which eco ds deduc ible paymen s o an Indi idual Pension
Scheme (IPS), o TR 4.5.1, which eco ds capi al in an Indi idual Pension Accoun (IPA).
Kispells, o which he measu ed du a ion o spells is ζi={ζi
0,ζi
1,...,ζi
Ki}. No e ha mea-
su ed du a ion is no necessa ily equal o he ue leng h o he spell because o censo ing.
Assume ha he spells ζ=(ζ0,ζ1,...,ζK) a e d awn om a p opo ional haza ds model wi h
a baseline haza d b0. De ining
[b]
d1,d2(ζ;b)≡X
(j,k):1≤j≤k≤K
(bd2
1
ζj=d1,ζk≥d2−bd1
1
ζj=d2,ζk≥d1)
hen
E
[ [b]
1, 2]=0∀¯
D≤d1<d2≤¯
Di and only i b=λb0 o some λ>0. This gi es ˜
D(˜
D+1)
2
momen condi ions, whe e ˜
D≡¯
D−¯
D. I is impo an o no e ha unde his p ocedu e, we
eco e he baseline haza ds bup o a mul iplica i e cons an , and so we no malize b1=1.
To es ima e b0:
ˆ
b0=a gmin
b¡1
N
N
X
i=1
[b]
d1,d2(ζi;b)¢TW¡1
N
N
X
i=1
[b]
d1,d2(ζi;b)¢
whe e Wis a posi i e de ini e weigh ing ma ix. We use wo-s ep easible GMM à la Hansen
(1982). In he i s s ep, we use he iden i y ma ix as he weigh ing ma ix. In he second
s ep, we ake ou es ima es om he i s s ep, b(1)
0, and use ˆ
W(ˆ
b0)−1as he weigh ing ma ix
7
in he second s ep whe e:7
ˆ
W(ˆ
b0)=³1
N
N
X
i=1
[b]
d1,d2(ζi;ˆ
b0) [b]
d1,d2(ζi;ˆ
b0)T´−1
The e a e se e al ad an ages o his app oach. Fi s , while Hono é (1993) p o ides con-
inuous ime iden i ica ion esul s o du a ion models wi h mul iple spells, he momen
condi ions used in he GMM es ima o a e based on disc e e ime iden i ica ion esul s. Sec-
ond, some app oaches ely on speci ica ion o a ail y dis ibu ion. Fo example, Nakamu a
and S einsson (2008) apply he empi ical model o Meye (1990) in hei analysis o p ice
spell du a ion and assume ha he ail y pa ame e ollows a gamma dis ibu ion o hei
baseline speci ica ion. Heckman and Singe (1984) no e ha misspeci ica ion o he ail y
dis ibu ion can bias he haza d unc ion. Ins ead, he app oach o Al a ez e al. (2021) im-
poses no es ic ions on he ail y dis ibu ion. Thi d, he GMM es ima o is consis en
when he numbe o indi iduals is la ge, and so i allows o a sho ime dimension. The
la e is impo an in ou se ing gi en ha we ely on annual da a co e ing 24 yea s.
E Al e na i e heo ies o pa icipa ion
E.1 Nons anda d p e e ences
Expec ed u ili y maximize s wi h s anda d p e e ences exhibi ing second-o de isk a e -
sion (e.g., CRRA u ili y) should always be willing o in es some money in s ocks as long
as he expec ed isk p emium is posi i e (Haliassos and Be au (1995)). This is because
such indi iduals a e e ec i ely isk neu al o small isks and isk has no i s -o de e -
ec . As such, a model whe e all agen s exhibi second-o de isk a e sion would need o
be augmen ed wi h addi ional ing edien s o mo i a e nonpa icipa ion, such as pa icipa-
ion cos s o backg ound isks. Some pape s ha e allowed o ime- a ying le els o isk
a e sion, wi h one popula me hod being o ha e habi - o ma ion p e e ences. Such p e -
e ences gene a e a nega i e ela ionship be ween weal h and isk a e sion.8Howe e , his
7Hansen (1982) show ha ˆ
W(ˆ
b0) con e ges in p obabili y o Ω≡
E
[ [b]
d1,d2(ζi;b0) [b]
d1,d2(ζi;b0)T] and ha W=
Ω−1is he mos e icien weigh ing ma ix.
8These s udies ha e ypically used habi - o ma ion p e e ences o help ep oduce empi ical pa e ns o
equi y p emia (e.g., Cons an inides (1990); Campbell and Coch ane (1999)).
8
will simply lead o ime- a ying isky asse sha es wi h no impac on he ex ensi e ma gin
o pa icipa ion as long as p e e ences s ill exhibi second-o de isk a e sion.9
To gene a e nonpa icipa ion exclusi ely h ough p e e ences, i s -o de isk a e sion is
needed (Segal and Spi ak (1990)).10 Unde such p e e ences, indi iduals ha e a kink in he
u ili y unc ion a some ce ain y poin , which can make isk a e sion locally in ini e and
ze o s ockholdings an op imal ou come. To gene a e dynamics in pa icipa ion, we would
need some agen s o exhibi ime- a ying i s -o de isk a e sion (Gomes e al. (2021)). In
addi ion, p e e ences would need o luc ua e a a easonably high equency o gene a e
sho - e m dynamics. Empi ical s udies ha e ypically ound posi i e and signi ican au-
oco ela ions in indi iduals’ isk p e e ences, sugges ing ha p e e ences a e mode a ely
s able, al hough co ela ions a e usually below 1 (Chuang and Schech e (2015); Dohmen
e al. (2016)).11
E.2 Risks aced by households
A s and o he li e a u e s udies how backg ound isks, pa icula ly labo income isk, can
a ec po olio alloca ions. Theo e ically, he impac o labo income isk depends on he
na u e o he isk (Vissing-Jø gensen (2002)). Fi s , i labo income is iskless, his should
lead o a highe in es men in isky inancial asse s because such labo income is e ec i ely
equi alen o holding a iskless bond. Second, i labo income is isky bu unco ela ed wi h
s ock e u ns, hen indi iduals should il hei po olio away om s ocks, as he e is al-
eady isk coming om human weal h.12 Thi d, i labo income is isky and co ela ed wi h
s ock e u ns, hen he e is a hedging componen ha uns wi h he opposi e sign o he co -
ela ion. Fo example, i business cycle isk p oduces a posi i e co ela ion be ween labo
9B unne meie and Nagel (2008) empi ically es whe he weal h luc ua ions a ec isky asse sha es and
ind no clea ela ionship, which hey a gue lends suppo o a CRRA model o e a model wi h habi - o ma ion
p e e ences.
10A ange o p e e ences exis ha exhibi i s -o de isk a e sion including, bu no limi ed o p ospec he-
o y (Kahneman and T e sky (1979)), ank-dependen expec ed u ili y (Quiggin (1982); Eps ein and Zin (1990)),
disappoin men a e sion (Gul (1991); Ang e al. (2005); Rou ledge and Zin (2010)), news u ili y (Pagel (2018)),
and ambigui y a e sion (Gilboa and Schmeidle (1989); Cao e al. (2005)).
11Pa o hese impe ec co ela ions could e lec measu emen e o (Schildbe g-Hö isch (2018)).
12Fage eng e al. (2017b) s udies he impac o uninsu able wage isk on po olio sha es using No wegian
da a. They ind a signi ican ma ginal e ec o such isk on po olio sha es, al hough he economic impac is
limi ed because he size o his wage isk is small. Vissing-Jø gensen (2002) inds a nega i e impac o non i-
nancial income ola ili y on bo h he p obabili y o s ock ma ke pa icipa ion and he p opo ion o weal h
in es ed in s ocks condi ional on pa icipa ing.
9
Figu e H2: Spell leng h dis ibu ion a he household le el
0 .05 .1
F ac ion o spells (n=1230745)
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22
Leng h o spell (in yea s)
No es. This his og am plo s he p opo ion o spells o di e en leng hs in he No wegian da a based on
he household-le el balance shee . A household is ea ed as pa icipa ing in he s ock ma ke in yea i
household-le el holdings o public equi y exceeds $150. We ake all spells beginning a any poin om 1994-
2013. The x-axis gi es he spell leng h (in yea s) and he y-axis shows he p opo ion o spells belonging o a
pa icula spell leng h. Righ -censo ed spells a e excluded om he plo , bu a e used in calcula ing p opo -
ions.
16
Figu e H3: Spell leng h dis ibu ion ( obus ness o gi s/inhe i ance)
(a) No gi abo e 10,000 NOK
0 .05 .1 .15
F ac ion o spells (n=1728114)
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22
Leng h o spell (in yea s)
(b) No (g and)pa en dea h
0 .05 .1 .15
F ac ion o spells (n=1652256)
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22
Leng h o spell (in yea s)
(c) No (g and)pa en pa icipa ion
0 .05 .1 .15
F ac ion o spells (n=1214771)
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22
Leng h o spell (in yea s)
No es. This his og am plo s he p opo ion o spells o di e en leng hs in he No wegian da a o di e en
subsamples in ended o deal wi h conce ns ha sho spells a e d i en by gi s and inhe i ances. Fo all panels,
we ake spells beginning a any poin om 1994-2013. The x-axis gi es he spell leng h (in yea s) and he y-
axis shows he p opo ion o spells belonging o a pa icula spell leng h. Righ -censo ed spells a e excluded
om he plo , bu a e used in calcula ing p opo ions. Panel (a) excludes all indi iduals who ecei e a gi
o inhe i ance abo e 10,000 NOK (based on ax eco ds) in he yea o o be o e en y. Panel (b) excludes all
en an s who expe ience he dea h o a pa en o g andpa en in he yea o o be o e en y. Panel (c) excludes
all en an s o whom a pa en o g andpa en pa icipa ed in he yea o o be o e en y.
17
Figu e H4: Spell leng h dis ibu ion excluding employee s ocks
0 .05 .1 .15
F ac ion o spells (n=561627)
1 2 3 4 5 6 7 8 9 10 11
Leng h o spell (in yea s)
No es. This his og am plo s he p opo ion o spells o di e en leng hs in he No wegian da a excluding en-
an s who hold s ocks in he company hey wo k o . Such indi iduals a e iden i ied using he Sha eholde
Regis y and demog aphic in o ma ion abou place o wo k. We ake all spells beginning a any poin om
2004-2013 (Sha eholde egis y da a begin in 2004). The x-axis gi es he spell leng h (in yea s) and he y-axis
shows he p opo ion o spells belonging o a pa icula spell leng h. Righ -censo ed spells a e excluded om
he plo , bu a e used in calcula ing p opo ions.
Figu e H5: Spell leng h dis ibu ion excluding small in es o s
(a) In es >$500
0 .05 .1 .15
F ac ion o spells (n=1555209)
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22
Leng h o spell (in yea s)
(b) In es >$1000
0 .05 .1 .15
F ac ion o spells (n=1333614)
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22
Leng h o spell (in yea s)
No es. This his og am plo s he p opo ion o spells o di e en leng hs in he No wegian da a excluding en-
an s who in es a “small” amoun o money a he poin o en y. The le panel only uses indi iduals who
in es a leas $500 a he poin o en y, while he igh panel equi es an in es men o a leas $1,000. Fo
bo h panels, we ake spells beginning a any poin om 1994-2013. The x-axis gi es he spell leng h (in yea s)
and he y-axis shows he p opo ion o spells belonging o a pa icula spell leng h. Righ -censo ed spells a e
excluded om he plo , bu a e used in calcula ing p opo ions.
18
Figu e H6: Spell leng h dis ibu ion using only i s spells
0 .05 .1 .15
F ac ion o spells (n=1407996)
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22
Leng h o spell (in yea s)
No es. This his og am plo s he p opo ion o spells o di e en leng hs in he No wegian da a using only he
i s eco ded spell o a gi en pa icipan . We ake all i s spells beginning a any poin om 1994-2013. The
x-axis gi es he spell leng h (in yea s) and he y-axis shows he p opo ion o spells belonging o a pa icula
spell leng h. Righ -censo ed spells a e excluded om he plo , bu a e used in calcula ing p opo ions.
Figu e H7: Weal h dis ibu ion by spell leng h
0 .05 .1 .15
F ac ion o indi iduals
1 2 3 4 5 6 7 8 9 10
Weal h decile a poin o en y
>2 yea s ≤2 yea s
No es. This igu e plo s he p opo ion o s ock ma ke pa icipan s belonging o di e en weal h deciles as
measu ed a he poin o en y, sepa a ely o sho spelle s (pa icipa e o ≤2 yea s) and longe - e m pa ici-
pan s (>2 yea s).
19
Figu e H8: Impac o age on he p obabili y o a sho spell
-.03 -.02 -.01 0
Coe icien es ima e
20-29 30-39 40-49 50-59 60-69 70+
Age g oup
No es. This igu e plo s he coe icien es ima es o he ixed e ec s on age ollowing es ima ion o Equa ion 1.
Age is measu ed a he poin o en y and indi iduals a e g ouped in o 10-yea bins. 95% con idence in e als
a e shown. The ed line ep esen s a null ela i e e ec .
20
Figu e H9: Spell leng h dis ibu ion by obse able cha ac e is ics
(a) Income
0 .05 .1 .15
F ac ion o spells (n=1803107)
0 5 10 15 20
Leng h o spell (in yea s)
Below median Abo e median
(b) Financial weal h
0 .05 .1 .15
F ac ion o spells (n=1822414)
0 5 10 15 20
Leng h o spell (in yea s)
Below median Abo e median
(c) Educa ion
0 .05 .1 .15
F ac ion o spells (n=1808241)
0 5 10 15 20
Leng h o spell (in yea s)
No deg ee Deg ee
(d) Gende
0 .05 .1 .15
F ac ion o spells (n=1822438)
0 5 10 15 20
Leng h o spell (in yea s)
Female Male
(e) By asse class
0 .05 .1 .15 .2
F ac ion o spells (n=1158139)
0 5 10 15 20
Leng h o spell (in yea s)
En e s only in o unds En e s only in o s ocks
No es. This his og am plo s he p opo ion o spells o di e en leng hs in he No wegian da a o di e en
obse able cha ac e is ics, namely income (a), weal h (b), educa ion (c), and gende (d). Panel (e) looks a
indi iduals who en e in o mu ual unds s. di ec ly held s ocks. Fo his panel, we exclude hose en an s
who choose o in es in bo h a he poin o en y. Fo panels (a) o (d), we ake all spells beginning a any
poin om 1994-2013, while o panel (e) we use spells s a ing om 1999-2013 as we canno sepa a e und
and s ock holdings in he ax da a un il 1998. The x-axis gi es he spell leng h (in yea s), and he y-axis shows
he p opo ion o spells belonging o a pa icula spell leng h. Righ -censo ed spells a e excluded om he
plo , bu a e used in calcula ing p opo ions.
21
Figu e H10: P e alence o liquidi y shocks in exi yea by spell leng h
0 .05 .1 .15
P opensi y o ace a liquidi y shock
0 1 2 3 4 5 6 7 8 9 10 >10
Spell leng h
No es. This igu e shows he p opo ion o exi e s o di e en spell leng hs expe iencing a leas one o ou
po en ial liquidi y needs in he exi yea . The ou shocks conside ed a e buying a house (obse ed in housing
ansac ions da a), di o ce, unemploymen (in e ed h ough eceip o unemploymen bene i s), and a la ge
all in income o >50%. The a -le ba (spell leng h o ze o) gi es he p e alence o liquidi y shocks o e
nonexi obse a ions (i.e. con inuing pa icipan s). The a - igh ba g oups all exi e s wi h spell leng hs abo e
10 yea s. 95% con idence in e als a e shown.
Figu e H11: P e alence o sho spells o e ime
0 .1 .2 .3 .4
P opo ion o yea -2 en an s exi ing by
1995 2000 2005 2010 2015
Yea
No es. This igu e plo s he p opo ion o yea −2 en an s who lea e he s ock ma ke by yea . The shaded
a eas a e s ock ma ke down u n yea s in which he Oslo Bø s Benchma k Index ell by a leas 10%.
22
Figu e H12: Cox p opo ional haza d unc ion o exi om pa icipa ion
0 .2 .4 .6 .8 1
Haza d a e (no malized)
0 5 10 15
Time since en y
Al a ez Cox
No es. This igu e plo s he baseline haza d o exi om pa icipa ion es ima ed using a Cox p opo ional
haza ds model, which akes he o m:
hi(d)=exp(Xiβ)bd
whe e Xiis a se o indi idual cha ac e is ics and bdis he baseline haza d. Xicon ains he same con ols as
in Table 2. The baseline haza d es ima ed using he Al a ez e al. (2021) me hodology (see Sec ion 3.1.2) is also
shown. To acili a e compa ison wi h he Al a ez e al. (2021) haza d unc ion, he Cox baseline haza d has
been no malized o 1 a du a ion d=1.
Figu e H13: Decomposi ion o en y a e in o een an s and new en an s
0 .5 1 1.5 2 2.5
En y a e (%)
2010
2011
2012
2013
2014
2015
2016
Re-en an New en an
No es. This igu e decomposes he s ock ma ke en y a e in a gi en yea in o wo componen s: en y by
o me pa icipan s (“Re-en an ”) and new en an s who ha e no pa icipa ed be o e. The en y a e in yea
is he p opo ion o nonpa icipan s in yea −1 who en e in yea .
23
Figu e H14: Reen y in o di e en asse classes by p e ious asse class choice
0 .2 .4 .6 .8
P opo ion o e-en an s
Funds be o e S ocks be o e
En e s in o unds En e s in o s ocks
En e s in o bo h
No es. This igu e plo s he p opo ion o een an s going in o unds, s ocks o bo h by he choice o unds s.
s ocks in hei p e ious spell.
Figu e H15: Reen y a e o e ime
0 .2 .4 .6 .8
Reen y a e
1995 2000 2005 2010 2015
Yea
Reen e s wi hin 4 yea s Oslo s ock ma ke down u n
No es. This igu e plo s he p opo ion o exi e s o a gi en yea who een e wi hin he nex 4 yea s. The shaded
a eas a e s ock ma ke down u n yea s in which he Oslo Bø s Benchma k Index ell by a leas 10%.
24
Figu e H16: Impac o income and weal h on he p obabili y o een y
(a) Income
0 .05 .1
Coe icien es ima e
1 2 3 4 5 6 7 8 9 10
Income decile
(b) Financial weal h
-.05 0 .05 .1 .15
Coe icien es ima e
1 2 3 4 5 6 7 8 9 10
Financial weal h decile
No es. This igu e plo s he coe icien es ima es o he ixed e ec s on income (A) and weal h (B) deciles
ollowing he es ima ion o Equa ion 2. Va iables a e measu ed a he poin o exi , and deciles a e based on
he ull No wegian popula ion aged 20 and abo e in ha yea . The e ec s a e es ima ed ela i e o he median
g oup. 95% con idence in e als a e shown. The ed line ep esen s a null ela i e e ec .
Figu e H17: Impac o age on he p obabili y o een y
-.25 -.2 -.15 -.1 -.05 0
Coe icien es ima e
20-29 30-39 40-49 50-59 60-69 70+
Age g oup
No es. This igu e plo s he coe icien es ima es o he age g oup ixed e ec s ollowing es ima ion o Equa ion
2. Age is measu ed a he poin o exi , and indi iduals a e g ouped in o 10-yea bins. 95% con idence in e als
a e shown. The ed line ep esen s a null ela i e e ec .
25
Figu e H28: A e age sa e inancial asse g ow h a ound en y/exi (1 yea spelle s)
(a) Model
-5 -4 -3 -2 -1 0 1 2 3
Yea s since exi
-0.06
-0.04
-0.02
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
Sa e inancial asse s (annual g ow h)
(b) Da a
-.2 -.1 0 .1 .2
Sa e inancial asse s (annual g ow h ela i e o -5)
-5 -4 -3 -2 -1 0 1 2 3
Yea s since exi
No es. This igu e plo s he e olu ion o sa e inancial asse g ow h a ound exi o people who had a 1-yea
spell in he simula ed model wi h EBL (panel a) and in he No wegian da a (panel b). Fo he model, we look
a he i s s ock ma ke exi e en o each agen and plo he a e age g ow h o sa e inancial asse holdings in
each yea , anging om 5 yea s be o e he exi o 3 yea s a e he exi . Fo he da a, we also ake all indi iduals
o whom he i s spell las ed 1 yea . The igu e plo s he βhcoe icien es ima es om he ollowing eg ession:
yi =αi+γ +
3
X
h=−5
βh
1
{ =Ei+h}+ϵi
whe e yi is he annual g ow h a e o sa e inancial asse holdings, αia e indi idual ixed e ec s, and γ a e
yea ixed e ec s. Eideno es he yea in which indi idual iexi s he s ock ma ke , and
1
{ =Ei+h} is an
indica o a iable equal o 1 i he yea equals hyea s a e he exi e en . The h= −5 case is omi ed, and so
coe icien s ell us he g ow h a e in yea ela i e o yea Ei−5. G ow h a es a e immed a he 5 h and 95 h
pe cen iles o each yea p io o es ima ion, and we es ic a en ion o hose who in es a leas $5,000 in he
s ock ma ke . Fo bo h panels, he do ed e ical lines e lec he pe iods jus be o e en y and exi (one yea
la e ).
32
Figu e H29: A e age sa ings a e a ound en y/exi (1 yea spelle s) in model wi h belie s
-5 -4 -3 -2 -1 0 1 2 3
Yea s since exi
0.35
0.4
0.45
0.5
Sa ing a e
No es. This igu e plo s he e olu ion o he sa ings a e ( o al sa ings di ided by cash on hand) o people who
had a 1-yea spell in he simula ed model wi h EBL. We look a he i s s ock ma ke exi e en o each agen
and plo he a e age sa ings a e in each yea , anging om 5 yea s be o e he exi o 3 yea s a e he exi . The
do ed e ical lines e lec he pe iods jus be o e en y and exi (one yea la e ).
Figu e H30: P e alence o p i a e pensions amongs exi e s by spell leng h
0 .05 .1 .15 .2
P opo ion o indi iduals who a e p i a e pension con ibu o s
0 1 2 3 4 5 6 7 8 9 10 >10
Spell leng h
No es. This igu e shows he p opo ion o exi e s o di e en spell leng hs pa icipa ing in p i a e pension
accoun s as o hei exi yea . An indi idual is said o be pa icipa ing in p i a e pensions in hei exi yea i hey
pu money in o a p i a e pension ei he in he cu en yea o in a pas yea . Pa icipa ion has occu ed i ei he
o he ollowing wo i ems in he ax eco ds is nonze o: TR 3.3.5, which eco ds deduc ible paymen s o an
Indi idual Pension Scheme (IPS), o TR 4.5.1, which eco ds capi al in an Indi idual Pension Accoun (IPA). The
a -le ba (spell leng h o ze o) gi es he p e alence o p i a e pensions shocks o e nonexi obse a ions (i.e.
con inuing pa icipan s). The a - igh ba g oups all exi e s wi h spell leng hs abo e 10 yea s. 95% con idence
in e als a e shown.
33
Figu e H31: Spell leng h dis ibu ion excluding indi iduals wi h a p i a e pension accoun
0 .05 .1 .15
F ac ion o spells (n=1822438)
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22
Leng h o spell (in yea s)
No es. This his og am plo s he p opo ion o spells o di e en leng hs in he No wegian da a excluding in-
di iduals who a any poin in he sample hold a p i a e pension accoun . Pa icipa ion has occu ed i ei he
o he ollowing wo i ems in he ax eco ds is nonze o: TR 3.3.5, which eco ds deduc ible paymen s o an
Indi idual Pension Scheme (IPS), o TR 4.5.1, which gi es capi al in an Indi idual Pension Accoun (IPA). The
x-axis gi es he spell leng h (in yea s) and he y-axis shows he p opo ion o spells belonging o a pa icula
spell leng h. The a - igh ba gi es he p opo ion o hese spells ha a e igh censo ed.
34
Re e ences
AGNEW, J., P. BALDUZZI,AND A. SUNDÉN (2003): “Po olio Choice and T ading in a La ge 401(k)
Plan,” Ame ican Economic Re iew, 93(1), 193–215.
ALVAREZ, F. E., K. BOROVI ˇ
CKOVÁ,AND R. SHIMER (2021): “Consis en E idence on Du a ion Depen-
dence o P ice Changes,” Wo king Pape 29112, Na ional Bu eau o Economic Resea ch.
AMERIKS, J., AND S. P. ZELDES (2004): “How do household po olio sha es a y wi h age,” Discussion
pape , Columbia Uni e si y.
ANG, A., G. BEKAERT,AND J. LIU (2005): “Why s ocks may disappoin ,” Jou nal o Financial Eco-
nomics, 76(3), 471–508.
BRANDSAAS, E. E. (2021): “Household S ock Ma ke Pa icipa ion and Exi : The Role o Homeowne -
ship,” Wo king pape .
BROWN, J. R., Z. IVKOVI ´
C, P. A. SMITH,AND S. WEISBENNER (2008): “Neighbo s Ma e : Causal Com-
muni y E ec s and S ock Ma ke Pa icipa ion,” The Jou nal o Finance, 63(3), 1509–1531.
BRUNNERMEIER, M. K., AND S. NAGEL (2008): “Do Weal h Fluc ua ions Gene a e Time-Va ying Risk
A e sion? Mic o-e idence on Indi iduals,” Ame ican Economic Re iew, 98(3), 713–36.
CALVET, L. E., J. Y. CAMPBELL,AND P. SODINI (2009): “Figh o Fligh ? Po olio Rebalancing by
Indi idual In es o s,” The Qua e ly Jou nal o Economics, 124(1), 301–348.
CAMPBELL, J. Y., AND J. H. COCHRANE (1999): “By Fo ce o Habi : A Consump ion-Based Explana ion
o Agg ega e S ock Ma ke Beha io ,” Jou nal o Poli ical Economy, 107(2), 205–251.
CAO, H. H., T. WANG,AND H. H. ZHANG (2005): “Model Unce ain y, Limi ed Ma ke Pa icipa ion,
and Asse P ices,” The Re iew o Financial S udies, 18(4), 1219–1251.
CARROLL, C. D., B.-K. RHEE,AND C. RHEE (1994): “A e The e Cul u al E ec s on Sa ing? Some
C oss-Sec ional E idence,” The Qua e ly Jou nal o Economics, 109(3), 685–699.
(1999): “Does Cul u al O igin A ec Sa ing Beha io ? E idence om Immig an s,” Economic
De elopmen and Cul u al Change, 48(1), 33–50.
CHUANG, Y., AND L. SCHECHTER (2015): “S abili y o expe imen al and su ey measu es o isk, ime,
and social p e e ences: A e iew and some new esul s,” Jou nal o De elopmen Economics, 117,
151–170.
COCCO, J. F., F. J. GOMES,AND P. J. MAENHOUT (2005): “Consump ion and Po olio Choice o e he
Li e Cycle,” The Re iew o Financial S udies, 18(2), 491–533.
CONSTANTINIDES, G. (1990): “Habi Fo ma ion: A Resolu ion o he Equi y P emium Puzzle,” Jou nal
o Poli ical Economy, 98(3), 519–43.
DIETZ, P. O. (1968): “Componen s o a Measu emen Model: Ra e o Re u n, Risk, and Timing,” The
Jou nal o Finance, 23(2), 267–275.
DOHMEN, T., H. LEHMANN,AND N. PIGNATTI (2016): “Time- a ying indi idual isk a i udes o e
he G ea Recession: A compa ison o Ge many and Uk aine,” Jou nal o Compa a i e Economics,
44(1), 182–200.
35
EPSTEIN, L. G., AND S. E. ZIN (1990): “‘Fi s -o de ’ isk a e sion and he equi y p emium puzzle,”
Jou nal o Mone a y Economics, 26(3), 387–407.
FAGERENG, A., C. GOTTLIEB,AND L. GUISO (2017a): “Asse Ma ke Pa icipa ion and Po olio Choice
o e he Li e-Cycle,” The Jou nal o Finance, 72(2), 705–750.
FAGERENG, A., L. GUISO, D. MALACRINO,AND L. PISTAFERRI (2020): “He e ogenei y and Pe sis ence
in Re u ns o Weal h,” Econome ica, 88(1), 115–170.
FAGERENG, A., L. GUISO,AND L. PISTAFERRI (2017b): “Po olio Choices, Fi m Shocks, and Uninsu -
able Wage Risk,” The Re iew o Economic S udies, 85(1), 437–474.
FAGERENG, A., M. B. HOLM, B. MOLL,AND G. NATVIK (2019): “Sa ing Beha io Ac oss he Weal h
Dis ibu ion: The Impo ance o Capi al Gains,” Wo king Pape 26588, Na ional Bu eau o Eco-
nomic Resea ch.
FREDRIKSEN, D., AND E. HALVORSEN (2019): “Be egninge a pensjons o mue,” Discussion pape ,
S a is ics No way.
FREDRIKSEN, D., AND N. M. STØLEN (2018): “Re o m a o en lig jenes epensjon,” Discussion pape ,
S a is ics No way.
FUCHS-SCHÜNDELN, N., P. MASELLA,AND H. PAULE, PALUDKIEWICZ (2020): “Cul u al De e minan s
o Household Sa ing Beha io ,” Jou nal o Money, C edi and Banking, 52(5), 1035–1070.
GILBOA, I., AND D. SCHMEIDLER (1989): “Maxmin expec ed u ili y wi h non-unique p io ,” Jou nal o
Ma hema ical Economics, 18(2), 141–153.
GOMES, F., M. HALIASSOS,AND T. RAMADORAI (2021): “Household Finance,” Jou nal o Economic
Li e a u e, 59(3), 919–1000.
GUIN, B. (2017): “Cul u e and household sa ing,” Wo king Pape Se ies 2069, Eu opean Cen al Bank.
GUISO, L., P. SAPIENZA,AND L. ZINGALES (2003): “People’s opium? Religion and economic a i udes,”
Jou nal o Mone a y Economics, 50(1), 225–282.
(2006): “Does Cul u e A ec Economic Ou comes?,” Jou nal o Economic Pe spec i es, 20(2),
23–48.
GUL, F. (1991): “A Theo y o Disappoin men A e sion,” Econome ica, 59(3), 667–686.
HALIASSOS, M., AND C. C. BERTAUT (1995): “Why do so Few Hold S ocks?,” The Economic Jou nal,
105(432), 1110–1129.
HALIASSOS, M., T. JANSSON,AND Y. KARABULUT (2017): “Incompa ible Eu opean Pa ne s? Cul u al
P edisposi ions and Household Financial Beha io ,” Managemen Science, 63(11), 3780–3808.
HANSEN, L. P. (1982): “La ge Sample P ope ies o Gene alized Me hod o Momen s Es ima o s,”
Econome ica, 50(4), 1029–1054.
HECKMAN, J., AND B. SINGER (1984): “A Me hod o Minimizing he Impac o Dis ibu ional Assump-
ions in Econome ic Models o Du a ion Da a,” Econome ica, 52(2), 271–320.
HONG, H., J. D. KUBIK,AND J. C. STEIN (2004): “Social In e ac ion and S ock-Ma ke Pa icipa ion,”
The Jou nal o Finance, 59(1), 137–163.
36
HONORÉ, B. E. (1993): “Iden i ica ion Resul s o Du a ion Models wi h Mul iple Spells,” Re iew o
Economic S udies, 60(1), 241–46.
KAHNEMAN, D., AND A. TVERSKY (1979): “P ospec Theo y: An Analysis o Decision unde Risk,”
Econome ica, 47(2), 263–291.
KARLSSON, N., G. LOEWENSTEIN,AND D. SEPPI (2009): “The os ich e ec : Selec i e a en ion o
in o ma ion,” Jou nal o Risk and unce ain y, 38(2), 95–115.
KAUSTIA, M., AND S. KNÜPFER (2012): “Pee pe o mance and s ock ma ke en y,” Jou nal o Finan-
cial Economics, 104(2), 321–338.
MEYER, B. D. (1990): “Unemploymen Insu ance and Unemploymen Spells,” Econome ica, 58(4),
757–782.
NAKAMURA, E., AND J. STEINSSON (2008): “Fi e Fac s abou P ices: A Ree alua ion o Menu Cos
Models,” The Qua e ly Jou nal o Economics, 123(4), 1415–1464.
ODEAN, T. (1998): “A e In es o s Reluc an o Realize Thei Losses?,” The Jou nal o Finance, 53(5),
1775–1798.
OECD (2009): “OECD P i a e Pensions Ou look 2008,” Discussion pape , OECD.
OSILI, U. O., AND A. PAULSON (2008): “Wha Can We Lea n abou Financial Access om U.S. Im-
mig an s? The Role o Coun y o O igin Ins i u ions and Immig an Belie s,” The Wo ld Bank
Economic Re iew, 22(3), 431–455.
PAGEL, M. (2018): “A news-u ili y heo y o ina en ion and delega ion in po olio choice,” Econo-
me ica, 86(2), 491–522.
QUIGGIN, J. (1982): “A heo y o an icipa ed u ili y,” Jou nal o Economic Beha io & O ganiza ion,
3(4), 323–343.
ROUTLEDGE, B. R., AND S. E. ZIN (2010): “Gene alized Disappoin men A e sion and Asse P ices,”
The Jou nal o Finance, 65(4), 1303–1332.
SCHILDBERG-HÖRISCH, H. (2018): “A e Risk P e e ences S able?,” Jou nal o Economic Pe spec i es,
32(2), 135–54.
SEGAL, U., AND A. SPIVAK (1990): “Fi s o de e sus second o de isk a e sion,” Jou nal o Economic
Theo y, 51(1), 111–125.
SHILLER, R. J., S. FISCHER,AND B. M. FRIEDMAN (1984): “S ock P ices and Social Dynamics,” B ook-
ings Pape s on Economic Ac i i y, 1984(2), 457–510.
TAUCHEN, G., AND R. HUSSEY (1991): “Quad a u e-Based Me hods o Ob aining App oxima e Solu-
ions o Nonlinea Asse P icing Models,” Econome ica, 59(2), 371–396.
VISSING-JØRGENSEN, A. (2002): “Towa ds an Explana ion o Household Po olio Choice He e ogene-
i y: Non inancial Income and Pa icipa ion Cos S uc u es,” Wo king Pape 8884, Na ional Bu eau
o Economic Resea ch.
37
